New Titles

Impact of influence marketing on customer trusts

Khurram, Muhammad Khizar[22L-6333]

Submitted in fulfillment of the requirements for the degree of Bachelor of science of accounting and finance to the Department of Management sciences. Contents Abstract.. Acknowledgement 1.1 Background 1.2 Research Problem 1.3 Research Gaps 1.4 Objectives of the Study 1.5 Research Scope. 1.6 Literature Review 1.7 Conceptual Framework. 1.8 Theoratical Framework.. 1.9 Methodology 2.1 Analysis 2.2 Discussion 2.3. Managerial Implication and Future Needs References Questionnaire. Abstract The social media influencers are taking over the minds and buying behavior of Generation Z consumers in the skincare sector. The digital economy in Pakistan is developing at a high rate with skincare influencers on websites such as Instagram and YouTube particularly becoming quite widespread, but there is a general lack of consumer credibility in beauty products because of inaccurate information and unregulated promotion. Though the engagement and persuasiveness of the follower influencers can be improved by the presence of a feeling of friendship one-sidedness (i.e. follower-influencer par asocial bond), the interaction of influencer attributes and parasocial interaction with customer trust in Pakistani skincare brands remains unclear. To fill these gaps, this research study contends on how three aspects of influencer, such as expertise, content quality, and authenticity, affect the trust that Generation Z Pakistani skincare consumers have in the brands, and how a parasocial interaction mediates the connections between them. Our research design is a quantitative and cross-sectional survey: a pre-tested questionnaire (based on proven scales) was distributed to a sample population consisting of 300 Pakistani Gen Z, who subscribe to skincare influencers on a regular basis. Each attribute of the influencer was represented in the questionnaire as an adaptation of previous literature, as well as known Parasocial interaction and trust scales. The expertise, the quality of the content, and authenticity of the influencers, their Parasocial relationship with the influencer, and brand trust were rated by respondents using the 5-point Likert scales. The analysis of data was done through reliability analysis, multiple regression test and mediation test. Through the source credibility and Parasocial interaction theories, the greater the expert power of the influencer, the higher quality of content and authenticity, the more likely that consumer trust is positively predicted, and Parasocial interaction is partially mediating these relations. Such anticipated results would help explain major motivators of trust in influencer marketing of Pakistani skincare products, providing valuable information to marketers and guidance to research in the future.

Date Published:2026

An analytical study of organizational challenges and strategies responses in the outsourcing industry : Evidence from the case of stafflink

Yousaf, Muhammad [22L-6348]

Submitted in fulfillment of the requirements for the degree of Bachelor of accounting & finance to the Department of Management science. Table of Contents Executive Summary Acknowledgement List of Tables List of Figures List of Acronyms 1 Introduction 1.1 Overview of the Company - Stafflink Profile 1.2 Background and Establishment 2 Vision, Mission, and Core Values 2.2 Industry Position 2.3 Culture 2.4 Core Values 2.5 Company Size 2.6 Headquarters 2.7 Company Type 3 Organizational Structure 3.1 Organization 3.2 Services 3.3 Key Departments and Their Roles 3.4 Management Function 3.5 Work Environment 3.6 Marketing Campaign 3.7 Performance Growth 4 Scope of Study 4.1 Scope 5 Data Collection 5.1 Data. 5.2 Meeting 1: General Company Discussion 5.3 Meeting 2: Company Structure, Services & Workflow 5.4 Meeting 3: Company Problems & Strategic Challenges 6 Major Challenges Faced by Stafflink 6.1 Challenges 6.2 Talent Acquisition & Retention 6.3 Operational & Compliance 6.4 Project Delivery & Client Expectations 7 Limitations of the Study 7.1 Limitations 8. SWOT Analysis - Stafflink 8.1 Strengths 8.2 Weaknesses 8.3 Opportunities 8.4 Threats 9 Organizational Improvement Framework 9.1 Talent Acquisition and Retention 9.2 Operational Excellence & Compliance 9.3 Project Delivery and Client Expectations 10 Competitive Environment Analysis (Stafflink) 10.1 PESTLE Analysis 10.2 Porter's Five Forces Analysis 11 Teaching Note 11.1 Introduction 11.2 Case Analysis & Theoretical Frameworks 11.3 Reflection and Discussion 11.4 Teaching/Learning Objectives 12. Exhibits 12.1 Founder and Department Heads 12.2 Picture While Interviewing 13. References. Executive Summary In the first months of growth, this case study highlights how Stafflink responded to these challenges and how it avoided them. While the firm was starting as a staffing and HR service firm, Stafflink faced many of the first challenges faced by the company, including low brand recognition, inability to attract talented staff, and difficult competition from established companies. These deficiencies hindered clients and improved credibility in a saturated market. The problems with internal systems and technology, coupled with a lack of sufficient external systems and technological infrastructure, impeded hiring. While that process was in progress, Stafflink responded to the challenges with a structured and innovative approach. The organization increased its digital presence, expanded recruitment efforts, and adopted norms to help the internal divisions de-brick. To build trust, reduce errors, and provide more reliable services to the clients, data-based screening tools, improved communication practices, and enhanced operational practices helped the company gain more confidence, efficiency, and reliable services. Despite these improvements, Stafflink became a locally owned service provider, becoming a respected HR solutions provider that could offer services across markets. It is not surprising that this research study documented not only the problems the company was facing but the major changes that contributed to its development. Stafflink needed continuity and continuous learning, while focusing on centered service for clients. Keywords: staffing solutions, operational issues, recruitment processes, digital enhancement, organizational development, strategic problem-solving, HR industry.

Date Published:2026

The impact of hybrid work models on employ productivity and job satisfaction evidence from Pakistan service sector

Khwaja, Aayan[22L-6389]

Submitted in fulfillment of the requirements for the degree of Bachelor of science in Business analytics to the Department of Management sciences. Table of Contents Abstract. List of Tables. List of Figures.. 1. INTRODUCTION. 1.1 Background. 1.2 Hybrid Work Models. 1.3 Job Satisfaction 1.4 Work-Life Balance.. 1.5 The Research Problem 1.6 Research Questions. 1.7 Research Objectives. 1.8 Research Gap 1.9 Scope of the Study 2. LITERATURE REVIEW. 2.1 Hybrid Work Models. 2.2 Hybrid Work Models and Job Satisfaction 2.3 Hybrid Work Models and Work-Life Balance. 2.4 Work-Life Balance and Job Satisfaction 2.5 Mediating Role of Work-Life Balance 2.6 Moderating Role of Organizational Support. 2.7 Theoretical Framework. 2.8 Hypothesis Development.... 3. RESEARCH DESIGN AND METHODOLOGY 3.1 Research Design.. 3.2 Research Approach. 3.3 Research Strategy. 3.4 Data Collection Tool. 3.4.1. Data Collection Procedure. 3.4.2. Sample Size 3.4.3. Research Tool 3.4.4. Questionnaire Design and Measurement of Variables... 3.5 Population 3.6 Sampling Technique 3.7 Data Analysis Techniques. 3.7.1. Reliability Analysis 3.7.2. Correlation Analysis 3.7.3. Regression Analysis (Direct Effects: H1-H3). 3.7.4. Mediation Analysis (H4) 3.7.5. Moderation Analysis (H5). 3.7.6. Structural Equation Modeling (SEM). 3.8 Ethical Considerations 3.9 Data Analysis Plan. 3.9.1. Descriptive Statistics 3.9.2. Inferential Statistics. 3.10 Validity and Reliability 3.10.1. Reliability. 3.10.2. Construct Validity. 3.11 Limitations and Delimitations. 4. RESULTS 4.1 Descriptive Statistics. 4.2 Reliability Analysis (Alpha) 4.3 Indicators Reliability. 4.4 Internal Consistency. 4.5 Convergent Validity. 4.6 Discriminant Validity. 4.7 Validity Analysis 4.8 Construct Validity (KMO and Bartlett's Test). 4.9 Correlation Analysis 4.10 Regression Analysis. 4.11 Moderation Analysis 4.12 Hypotheses Testing Summary 5. DISCUSSION 6. Conclusion Theoretical and Practical Implications.... Limitations of the Study. Future Research Directions 6. REFERENCES. 7. APPENDIX. Abstract Rapid growth of hybrid workplaces has forever changed how organizations are structured in the Global Service Sector, including the Service Sector in Pakistan, due to a rapid transition to hybrid workplace models. Therefore, this study investigates the impact of Hybrid Work Models on Job Satisfaction (JS) for service sector employees while exploring the mediating effect of Work-Life Balance (WLB) and the moderating impact of Organizational Support (OS). The methodology of the study was quantitative, with a structured questionnaire administered to employees working in hybrid-work arrangements, which were sampled through purposive sampling across information technology (IT), Banking. Telecoms, Education, and Consulting organizations. The target sample size was 150-200 respondents in order to satisfy the minimum statistical requirement for the analyses performed in the study using regression and Structural Equation Modeling (SEM). The findings demonstrate a statistically significant and positive correlation between Hybrid Work Models (HWM) and Job Satisfaction. Additionally, the findings show that Work-Life Balance (WLB) has a positive impact on Job Satisfaction as well, therefore indicating that employees that have access to flexibility in their job will improve their ability to balance both their personal and work lives. The findings from the mediation analysis also indicate that WLB partially mediates the relationship between HWM and Job Satisfaction, therefore employees that work in hybrid work settings achieved greater levels of job satisfaction through WLB compared to those that do not have hybrid work arrangements. According to moderation testing findings, the Positive Impact of the Hybrid Work Model will be strengthened by Organizational Support; e.g., an organization will benefit more from a Hybrid Work Sy when providing adequate Resources, Communication and Technology Support of this study also represent a substantial addition to the growing body of evidence from Pakistan on Hybrid Working, with specific reference to the importance of Organizational Support to realizing the Positive Benefits of the Hybrid Work Model. Keywords: Hybrid Work Models (HWM), Work-Life Balance (WLB), Job Satisfaction, Organizational Support, Service Sector, Mediation, Moderation.

Date Published:2026

How E-Wallet adoption mediates the influence of behavioural biases on retail investment decision

Anwar, M. Khubaib [22L-6355]

Submitted in fulfillment of the requirements for the degree of Bachelor of Science in Accounting and Finance to the Department of Management Sciences. Abstract List of Figures List of Tables List of Acronyms 1. Introduction 1.1 Background 1.2 Problem Statement 1.3 Research Gap 1.4 Research Objective 1.5 Research Question 1.6 UN SDG's Goals 2. Literature Review 2.1 Retail Investment Decisions (RTD) 2.2 Behavioural Baises (BB) and Retail Investment decisions (RID) 2.3 E-wallet adoption (EWA), Behavioural biases (BB) and Retail Investment Decisions (RID) 2.4 Theoretical Framework 2.4.1 Independent Variable 2.4.2 Dependent Variable 2.4.3 Mediating Variable 2.5 Hypothesis Development 4. Research Methodology and Data Collection 4.1 Research Methodology 4.1.1 Purpose of Study 4.1.2 Type of study 4.1.3 Researcher's Interference & Data Collection 4.1.4 Unit of Analysis 4.1.5 Time Horizon 4.2 Data Collection 4.2.1 Measurement 4.2.2 Sampling 4.3 Data Analysis 4.4 Measuring Constructs 4.5 Assessment of Common Method Variance (CMV) 5. Results and Discussion 5.1 Exploratory Phase 5.2 Goodness of Data 5.3 Hypothesis Testing 5.4 Mediation testing 6. Conclusion 7. Practical Implications 8. Theoretical Implications 9. Limitations 10. Future Line of Research 11. References Appendix 6.1 Questionnaire. The rapid growth of financial technology has transformed how individuals manage, save, and invest their money, with e-wallets emerging as a central tool in digital financial ecosystems. This study examines how behavioural biases such as overconfidence, loss aversion, anchoring, and herding affect retail investment decisions, and investigates the mediating role of e-wallet adoption in this relationship. Although prior research has explored behavioural biases and fintech adoption independently, limited evidence exists on how e-wallet usage may alter or reinforce the influence of these biases, particularly in emerging economies like Pakistan where digital financial adoption is accelerating. Using a quantitative research design, the study collects data from retail investors who actively engage with e-wallet platforms such as Jazz Cash, Easypaisa, and Naya Pay. The findings highlight that behavioural biases significantly shape investment decision-making, often pushing investors toward irrational or emotionally driven choices. Moreover, results reveal that e-wallet adoption partially mediates this relationship by influencing how investors process information, perceive risk, and respond to market cues. Features such as real-time updates, transaction simplicity, and financial tracking tools can either amplify impulsive decisions or promote informed and disciplined investing. This research contributes to behavioural finance and fintech literature by providing a unified behavioural-technological model of retail investor behaviour. The study also offers practical insights for fintech companies, policymakers, and financial institutions to design interventions that improve financial decision-making, enhance investor protection, and promote sustainable digital financial inclusion. Keywords: Behavioural biases, Retail investment decisions, E-Wallet adoption, Overconfidence, Loss version, Herding, Anchoring.

Date Published:2026

Behavioral barriers to tax compliance in pakistan

Zainab, Meer [22L-6304]

Submitted in fulfillment of the requirements for the degree of Bachelor of Business analytics to the Department of Management science. Table of Contents 1. Introduction 1.1 Research Problem 1.2 Research Questions 1.3 Research Objectives 1.4 Research Gap 1.4 Scope of the Study 2. Literature Review 2.1 Attitude towards Tax Evasion 2.2 Perception of Government Spending and Its Influence on Attitudes Toward Tax Evasion 2.3 Social Norms and Peer Influence as Behavioral Determinants of Attitudes Toward Tax Evasion 2.4 Perception of Fairness as a Mediator of Attitudes Toward Tax Evasion 2.5 Theoretical Framework 3. Research Design and Methodology 3.1 Purpose of the Study 3.2 Type of Study 3.3 Extent of Researcher Interference 3.4 Study Setting 3.5 Unit of Analysis 3.6 Time Horizon 3.7 Data Source and Population 3.8 Sampling Technique and Data Collection Method 3.9 Ethical Considerations 3.10 Reliability and Validity of Measures 3.11 Data Collection 3.12 Sampling Technique and Representativeness 3.9 Challenges in Data Collection 4. Results and Analysis 4.1 Data Preparation and Variable Operationalization 4.2 Descriptive Statistics 4.3 Reliability Analysis 4.4 Pearson Correlation Analysis 4.5 Multiple Regression Analysis 4.6 Mediation Analysis 5. Discussion and Conclusion 5.1 Discussion of Findings 5.2 Policy Implications 5.3 Recommendations 5.4 Limitations 5.5 Future Research 5.6 Conclusion References. Abstract Pakistan continues to struggle with chronically low voluntary tax compliance despite repeated administrative reforms and digitization efforts, indicating that structural explanations alone cannot account for taxpayer behavior. Behavioral perspectives argue that compliance is shaped not only by deterrence but by taxpayers' perceptions of government performance, fairness, and social expectations. This study investigates the behavioral determinants of attitudes toward tax evasion in Pakistan by examining the roles of perceived government spending, social norms and peer influence, and perceived fairness. In particular, it assesses whether fairness serves as a mediation mechanism via which social and institutional cues influence people's views about tax evasion. The research uses secondary data from the World Values Survey (WVS) Wave 7, a nationally representative dataset collected across four major provinces of Pakistan. The study uses a cross-sectional and correlational design with little intervention from the researcher. Variable coding from verified WVS items, data extraction, descriptive analysis, and testing for hypotheses with statistical relationships and mediation logic are among the procedures. The study fills in important gaps in the literature, since Pakistani research has mostly concentrated on trust in institutions and corruption. By testing six directional hypotheses, this research evaluates both direct and indirect effects between the variables to understand how taxpayers form moral judgments about evasion. The results are intended to improve knowledge of Pakistani taxpayer psychology and offer evidence for creating more efficient, equitable compliance plans that go beyond administrative and enforcement measures. Keywords: tax compliance, tax evasion, tax morale, perception of government spending, social norms, peer influence, perceived fairness.

Date Published:2026

Examining the impact of internal deficiencies on financial losses in state- owned interprises

Umair, Abubakar [22L-6515]

Submitted in fulfillment of the requirements for the degree of Bachelor of science in Business analytics to the Department of Management sciences. Table of Contents Abstract. List of Tables List of Figures. List of Acronyms. 1. Introduction. 1.1 Background of the Study 1.2 Problem Statement 1.3 Research Gap 1.3.1 Minimal reliance on Audit-based Evidence. 1.3.2 Lack of Cross-Report Consolidation 1.3.3 Failure to consider Mediating Mechanisms 1.3.4 Summary of Research Gap. 1.4 Research Question 1.5 Research Objectives. 1.6 Scope of the Study 1.7 Significance of the Study 1.7.1 Theoretical Contribution 1.7.2 Methodological Innovation. 1.7.3 Policy and Reform Relevance.. 1.7.4 Audit System Strengthening 1.8 Organization of the Paper 2. Literature Review 3. 2.1 Public-Sector Enterprises and Performance Problems 2.2 Weaknesses in Governance in Enterprises 15 17 2.3 Financial Procurement Deficiencies 2.4 Service-Delivery and Operational Deficiencies.. 19 21 2.5 Monitoring Weaknesses and Managerial Response to Decline. 23 2.5.1 Monitoring Weaknesses in SOEs... 23 2.5.2 Mismatch of Attention in Monitoring Sectors Contributing to Organizational Decline 24 2.5.3 Monitoring Weaknesses as a Transmission Mechanism.... 25 2.6 Financial Distress and SOE Financial Losses. 26 2.6.1 Patterns of Financial Distress in Organizations 26 2.6.2 Financial Losses in State-Owned Enterprises. 27 2.6.3 Financial Losses as an Integrated Outcome of Internal Deficiencies. 2.7 Synthesis and Research Gap. 28 28 2.7.1 Synthesis of Empirical Evidence... 2.7.2 Identified Research Gaps. 8 Hypotheses Development. 2.8.1 Governance & Strategic Deficiencies - SOE Loss. 2.8.2 Governance & Strategic Deficiencies - Financial & Procurement Deficiencies.. 2.8.3 Financial & Procurement Deficiencies - Monitoring Weaknesses. 2.8.4 Financial & Procurement Deficiencies - Operational Deficiencies & Inefficien 2.8.5 Moderation of HR Deficiency: Governance & Strategic Deficiencies - Financ Procurement Deficiencies. 2.8.6 Moderation of HR Deficiency: Financial & Procurement Deficiencies - Operatio Deficiencies & Inefficiency... 2.8.7 Monitoring Weaknesses-SOE Loss.... 2.8.8 Financial & Procurement Deficiencies - SOE Loss 2.8.9 Operational Deficiencies - Inefficiency & SOE Loss. 3. Research Design and Methodology 3.1 Type of Study 3.2 Degree of Interference of Researcher 3.3 Study Setting. 3.4 Unit of Analysis 3.5 Time Horizon 3.6 Data Collection 3.6.1 Nature of Data Source. 3.6.2 Screening and Extraction. 3.6.3 Coding Process 3.6.4 Data Compilation. 3.7 Regression Model 4. Data Analysis and Results. 4.1 Data Preparation and Final Analytical Dataset. 4.2 Construction of Dependent Variable: SOE Loss Intensity 4.3 Construction of Independent Variables and Deficiency Blocks... 4.4 Descriptive Statistics. 4.5 Ordinal Regression Analysis. 4.6 Model Fitting and Overall Model Significance 4.7 Goodness-of-Fit and Pseudo R-square..... 58 58 4.8 Parameter Estimates and Predictor-Level Interpretation. 4.9 Test of Parallel Lines 59 62 4.10 Hypotheses Testing Summary 63 5. Discussion.... 66 5.1 Discussion of Main Empirical Findings. 5.2 Governance and Strategic Deficiency as a Predictor of SOE Loss 5.3 Financial and Procurement Deficiency as the Strongest Predictor. 5.4 Operational Deficiencies and Inefficiency as a Significant Loss Driver. 6 5.5 Monitoring Weakness: Theoretical Importance but Non-significant Direct Effect...... 5.6 HR Deficiency: Moderation in Theory and Main-effect Limitation in Current Model 5.7 Alignment with Existing Literature 5.8 Implications for SOE Reform in Pakistan 5. Conclusion and Implications..... 6.1 Conclusion 6.2 Theoretical Implications 6.3 Methodological Contribution. 6.4 Practical and Policy Implications. 6.5 Limitations of the Study. 6.6 Future Research Guidelines 6.7 Final Remarks References. Appendix A A.1 Purpose of the Tool. A.2 Input Requirements. A.3 Extraction Procedure. A.4 Output Structure. A.5 Quality Control Measures A.6 Final Output Appendix B B.1 Purpose of the Pipeline. B.2 Input Dataset and Software Requirements B.3 Data Cleaning and Preparation. B.4 Variable Remapping and Threshold-Based Coding. B.5 Dependent Variable Construction B.6 Regression-Ready Dataset Creation. B.7 Output Workbook Structure B.8 Validation Checks. Abstract This paper will discuss internal causes of inefficiency in the State-Owned Enterprises (SOEs) in Pakistan by looking at the systemic weaknesses reported in the Auditor General of Pakistan (AGP) Performance Audit Report (PARs). Based on a wide scope of the scrutiny of the previous literature on governance, organizational decline, financial distress, and monitoring systems, the study comes up with four main categories of internal weaknesses that persistently cripple the performance of the SOE including governance weaknesses, financial-procurement weaknesses, service-delivery weaknesses, and monitoring weaknesses. Based on these thematic implications, the study formulates a conceptual framework, which speculates that internal inadequacies have direct and indirect impacts, by way of monitoring weak points, on financial losses in SOEs. The research uses a quantitative content-analysis approach, transforming the observations of narratives audit into binary-coded variables that can be tested empirically. A cross-sectional dataset is comprised of PARs across years and sectors and a unit of analysis is an audit observation. This method gives a systematic and repeatable way of measuring evidence of the audit, a vital void in current Pakistani research on the public sector, traditionally based on descriptive or case-based evaluation. The discoveries point to the linking of internal deficiencies, thus, showing that governance failures undermine the checks and balances, financial-procurement anomalies pervert resource distribution and service-delivery inefficiencies reduce operational ability. Detection weaknesses become both a consequence of the deficiencies and an intermediary force that increases the effects on financial losses. The research has important policy implications in SOE reform, including the necessity to have multi-pronged interventions that would reinforce governance, increase procurement integrity, operational efficiency, and consolidate effective monitoring systems. In general, this study furnishes a theoretically based and empirically knowledgeable background to comprehend and manage endemic performance problems in the field of the SOE sector in Pakistan. Keywords: Pakistan, State-Owned Enterprises (SOEs), governance deficiencies, financial-procurement weaknesses, service-delivery performance, monitoring weaknesses, HR-capacity, financial losses, Auditor General of Pakistan.

Date Published:2026

Financial health forecasting using multi-model machine learning and economic indicators

Umer, Muhammad [22L-6509]

Submitted in fulfillment of the requirements for the degree of Bachelor of science in Business analytics to the Department of Management sciences. Table of Contents Abstract Acknowledgements. Table of Contents. List of Tables. List of Figures. List of Acronyms 1. Introduction. 1.1. Problem Statement. 1.1.1. Objectives 1.1.2. Interests and Motivation 1.1.3. Scope of the Study 1.1.4. Significance of the Research.. 1.1.5. Impact Statements. 1.2. Identification of Gaps and Opportunities. 1.2.1. Research Gaps in Financial Prediction Analytics. 1.2.2. Market Gaps in Emerging Markets. 1.2.3. Opportunities for Further Research and System Development 1.3. Brief Discussion About Approach and Methodology. 1.3.1. Overview of the Approach. 1.3.2. Stages of Methodology 1.3.3. Dataset Description 1.3.4. Utility of the Dataset in Addressing the Problem 1.3.5. Summary of Dataset Utility 1.4. Alignment of the Project with Sustainable Development Goals (SDGs) 2. Literature Review. 2.1. Review of Previous Literature 2.1.1. Ensembled Direct Multi-Step Forecasting Methodology with Comparison on Macroeconomic and Financial Data 2.1.2. Machine Learning and Deep Learning in Computational Finance: A Systematic Review 2.1.3. Financial Distress Prediction for Small and Medium Enterprises Using Machine Learning Techniques. 2.1.4. Survey, Classification, and Critical Analysis of the Literature on Corporate Bankruptcy and Financial Distress Prediction (2024). 2.1.5. Financial Applications of Machine Learning: A Literature Review (2023). 2.1.6. Macroeconomic Forecasting Using Factor Models and Machine Learning: An Application to Japan (2020) 2.1.7. Machine Learning in Business and Finance: A Literature Review and Research Opportunities (2024). 2.1.8. Predictive Modeling for the Moroccan Financial Market: A Nonlinear Time Series and Deep Learning Approach (2025) 2.1.9. Integrating Macroeconomic and Technical Indicators in Stock Market Forecasting: Data-Driven Approach (2025). 2.1.10. Hybrid Machine Learning Models for Long-Term Stock Market Forecasting: Integrating Technical Indicators (2025). 2.1.11. Macro-Driven Stock Market Volatility Prediction: Insights from a New Hybrid Machine Learning Approach (2024). 2.1.12. Evaluating the Performance of Machine Learning Algorithms in Financial Market Forecasting: A Comprehensive Survey (2019). 2.1.13. Al-Driven Intelligent Financial Forecasting: A Comparative Study of Advanced Deep Learning Models for Long-Term Stock Market Prediction (2025). 2.1.14. How Well Do Machine Learning Models in Finance Work? (2025) 2.1.15. Global Stock Market Forecasting: Insights from Series and Parallel Combination o Machine Learning Models (2025) 2.2. Summary of Key Studies 2.2.1. Predictive Analytics and Decision Support 2.2.2. Multi-Ratio and Multi-Sector Forecasting. 2.2.3. Financial Ratio Models for Forecasting. 2.2.4. Integration of Macroeconomie Variables 2.2.5. Machine Learning vs. Deep Learning Approaches. 2.2.6. Hybrid and Ensemble Modeling. 2.2.7. Emerging Market Applications. 2.2.8. Long-Term and Multi-Step Forecasting 2.3. Major Findings 2.3.1. Performance of Ensemble and Hybrid Models 2.3.2. The Significance of Macro-Financial Integration. 2.3.3. Feature Selection and Data Preprocessing. 2.3.4. Multi-Step and Long-Term Forecasting 2.3.5. Comparative Performance: ML vs. DL Models 2.3.6. Sector-Specific Insights 2.3.7. Emerging Market Considerations 2.3.8. Practical Implications for Financial Forecasting 2.4. Methodologies of Past Research 2.4.1. Ensemble and Hybrid Modeling 2.4.2. Machine Learning Models. 2.4.3. Deep Learning Approaches. 2.4.4. Macroeconomic Integration and Factor Models 2.4.5. Feature Selection and Preprocessing Methods 2.4.6. Multi-Step and Multi-Output Forecasting 2.4.7. Comparative and Benchmarking Studies. 2.4.8. Hybrid Macro-Financial Modeling. 2.5. Implications for This Analysis. 2.5.1. Macro and Firm-Level Integration Reinforces Predictive Power. 2.5.2. Multi-Step and Multi-Ratio Forecasting for Realistic Decision Support. 2.5.3. The Most Promising Approach: Ensemble and Hybrid Modeling 2.5.4. Feature Engineering and Ratio Stability Across Sectors. 2.6. Gaps and Opportunities. 2.6.1. Limited Focus on Emerging Markets (Especially Pakistan). 2.6.2. Absence of Multi-Output Financial Ratio Forecasting. 2.6.3. Lack of Integration of Macro and Microeconomic Indicators.. 2.6.4. Inconsistent Reporting Standards and Missing Data in Pakistan. 2.6.5. Lack of Decision-Support Tools for Non-Expert Investors.. 2.6.6. Sector-Specific Behavior Not Modeled. 3. Methodology. 3.1. System Architecture Overview. 3.2. Web Portal Description. 3.3. Backend Description 3.4. Data Storage Process. 3.5. Authentication Mechanism 3.6. Automated Excel File Handling. 3.7. Database Schema Overview 3.8. Data Processing Pipeline. 3.8.1. Initial File Submission and Validation. 3.8.2. Extracting and Cleaning Financial Data 3.8.3. Automated Ratio Calculation and Storage. 3.8.4. Final Verification and Data Readiness 3.9. Deployment and Hosting on AWS 3.10. Web Portal Front-End Images 3.11. Proposed Analytical Techniques.. 3.11.1. Supervised Machine Learning Models 3.11.2. Feature Engineering Using Financial Ratios 3.11.3. Integration of External Macroeconomic Variables. 3.11.4. Feature Selection and Dimensionality Reduction. 3.11.5. Data Normalization and Scaling 3.11.6. Evaluation Techniques and Error Analysis. 3.11.7. Cross-Validation and Model Training Flow 3.11.8. Justification of Selected Techniques. 3.12. Novelty Statement and Unique Contributions 3.13. Research Implications and Broader Impact 3.14. Comparison with Previous Research 3.15. Benefits to Stakeholders from Advanced Financial Modeling. 3.15.1. Improved Decision-Making for Investors. 3.15.2. Support for Financial Institutions 3.15.3. Economic Development and Stability 3.15.4. Broader Societal Impacts 3.15.5. Enhancing Accessibility and Education.. 3.16. Packages Required 3.16.1. Packages Used in Web Portal 3.16.2. Packages Used for Data Cleaning and Preparation. 3.16.3. Packages Used for Model Training, Evaluation, and Visualization 3.17. Suppression of Messages and Warnings. 3.17.1. Handling Messages. 3.17.2. Suppression Techniques. 3.17.3. Implementation Considerations 3.18. Data Preparation. 3.18.1. Original Data Overview 3.18.2. Handling Missing Values. 3.18.3. Data Cleaning Process 3.19. Final Dataset 3.20. Data Privacy Guidelines 3.20.1. Protection and Anonymization of Sensitive Data. 3.20.2. Avoiding Bias in Data Collection and Analysis 3.20.3 Transparency of Methods and Algorithms 3.20.4. Accuracy and Integrity of Data.... 3.20.5. Use of Data with Appropriate Permissions.. 4. Data Analysis 4.1. Exploratory Data Analysis. 4.1.1. Dataset Overview and Descriptive Statistics. 4.1.2. Sectoral Coverage and Distribution 4.1.3. Cross-Ratio Correlation Analysis 4.1.4. Temporal Trends in Key Ratios. 4.2. Macroeconomic Environment Analysis (2005-2022). 4.2.1. PKR/USD Exchange Rate Dynamics 4.2.2. Inflation and Consumer Price Index Trends 4.2.3. State Bank of Pakistan Policy Rate Trajectory 4.2.4. Global Oil Price Impact on Pakistani Corporates. 4.3. Sector-Wise Financial Profile and Analysis 4.3.1. Analysis of Financial Performance by Sector. 4.3.2. Portfolio Overview. 4.3.3. Market Performance Analysis. 5. Findings. 5.1. Overview of Models Used 5.2. Model Architectures. 5.2.1. XGBoost 5.2.2. Gradient Boosting Machine (GBM) 5.2.3. Extra Trees Regressor 5.2.4. Random Forest. 5.2.5. ElasticNet. 5.2.6. Lasso Regression 5.2.7. K-Nearest Neighbors (KNN) 5.3. Evaluation Metrics. 5.3.1. Metrics Used. 5.3.2. Justification of Metric Selection 5.4. Results and Discussion 5.4.1. Interpretation of Results.. 5.4.2. Comparative Insights Across Models 5.4.3. Graphical Representation of Models 5.4.4. Financial Health Score Analysis. 5.4.5. Sector-Level Analysis and Rankings. 5.4.6. Performance Analysis 6. Conclusion 6.1. Problem Statement Recap. 6.2. Methodology Recap. 6.2.1. Data and Methods 6.3. Key Insights and Findings 6.3.1. Findings... 6.4. Implications. 6.4.1. Impact on Stakeholders. 6.5. Limitations and Future Work. 6.5.1. Limitations 6.5.2. Future Improvements 6.6. Closing Remarks. 6.6.1. Final Insights. 6.6.2. Future Directions 7. References. Abstract The financial landscape of Pakistan's corporate sector is shaped by persistent macroeconomic volatility, heterogeneous industrial structures, and a chronic absence of forward-looking analytical tools accessible to the broader investment community. Traditional financial analysis methodologies predominantly grounded in historical ratio computation and retrospective balance-sheet assessment provide limited utility for anticipating changes in firm-level financial health, particularly in a market environment characterized by frequent exchange rate shocks, inflation oscillations, and monetary policy rate cycles. This project develops, implements, and empirically validates a comprehensive, automated predictive framework specifically designed to forecast the next-period financial ratios of companies listed on the Pakistan Stock Exchange (PSX) across fifteen distinct industrial sectors. The system rests on a cloud-based, automated financial data architecture hosted on Amazon Web Services, where authorized users upload company financial staternents in standardized Excel format through a secure Flask-based web portal. An automated pipeline validates, cleanses, restructures, and stores these statements in a centralized SQLite database, simultaneously generating a library of 34 financial ratios spanning profitability, liquidity, leverage, operational efficiency, investor performance, and asset quality. The predictive modeling phase applies seven rigorously evaluated machine learning algorithms XGBoost, Gradient Boosting, Extra Trees, Random Forest, Elastic Net, Lasso, and K-Nearest Neighbors trained on time-series feature sets incorporating lagged ratios, rolling statistics, year-over-year changes, and four macroeconomic variables: the PKR/USD exchange rate, CPI inflation, the State Bank of Pakistan policy rate, and global Brent crude oil prices. A five-fold time-series cross-validation protocol with a one-year forward gap is employed throughout to prevent look-ahead bias and ensure honest evaluation of out-of-sample predictive performance. Results confirm that 71 percent of the 34 target ratios achieve R-squared values exceeding 0.60 on held-out test data, with profitability and liquidity ratios consistently achieving the strongest prediction accuracy. XGBoost attained the highest average R-square of 0.655 across all ratios, while Lasso despite its comparative simplicity won the most individual ratio contests (11 of 34), demonstrating that regularized linear models remain highly competitive for ratios governed by approximately linear temporal dynamics. Market-based ratios proved materially harder to predict from fundamental accounting data alone a limitation explicitly attributed to investor sentiment dynamics and motivating the integration of Natural Language Processing-based sentiment analysis as a priority future development. A composite Financial Health Score (0-100), constructed from fourteen weighted ratios across profitability, liquidity, solvency, and efficiency dimensions, enables intuitive cross-company and cross-sector ranking. Analysis reveals that Oil and Gas, Technology, and Pharmaceutical/Chemical sectors lead in predicted 2023 financial health, while Sugar and Engineering/Steel sectors show the weakest overall conditions. The concentration of 69 percent of companies in the Grade C (Average) classification is interpreted as a direct consequence of the historically adverse macroeconomic conditions prevailing in 2022 record-high simultaneous inflation, exchange rate depreciation, and policy rate. Future development priorities include NLP-based sentiment analysis, Large Language Model integration for report summarization and interactive querying, real-time sector dashboards, expanded macroeconomic coverage, and deep learning multi-step forecasting. Keywords: financial health forecasting, machine learning, financial ratios, Pakistan Stock Exchange, macroeconomic integration, sector-wise analysis, composite health score, XGBoost, Lasso, ensemble models, predictive analytics, emerging markets.

Date Published:2026

Prodex-from fragmented HR data to unified intelligence an applied research study on the design and evaluation of multi agent framework for HR analytics in the technology sector

Nawaz, Husnain[22L-8445]

Submitted in fulfillment of the requirements for the degree of Bachelor of science in Business analytics to the Department of Management sciences. Contents Abstract Acknowledgement Contents List of Tables: List of Figures. 1. Introduction. 1.1. Background and Context... 1.2. Development of HR Analytics: The Intelligence Shift. 1.3. The Tech Efficiency Paradigm Shift: The Post 2022 grid. 1.4. The Virtual Workplace and the Surplus of Digital Exhaustion. 1.5. The Effect of Generative Al and Multi-Agents Architectures.. 1.6. Problem Statement. 1.7. The Generating Contention: Semantic Fragmentation and Data Silos. 1.8. Analytical Gap: Weaknesses of Current Solutions. 1.9. Managerial Consequence Latency and Reactive Management 1.10. The security paradox: access vs Governance. 1.11. Research Objectives. 1.12. Scope and Delimitations 1.13. Scope of the Study 1.14. Delimitations.. 1.15. Significance of the Study 1.16. Academic Significance. 1.17. Industrial Significance 1.18. Factual Issues and technical problems. 1.19. Limitations of the Study. 2. Literature Review. 2.1. Workforce Analytics: 2.2. Taylorism vs. Agile: The Scientific Management to Adaptive Autonomy. 2.3. The Peril of Proxy Metrics and Goodhart Law. 2.4. The Data Fragmentation Problems in High-dimensional situations. 2.5. Static vs. Dynamic Organizational Reality Knowledge Cutoffs. 2.6. The Retrieval-Augmented Generation (RAG) Theoretical Mechanics. 2.7. Vector Embeddings: Language to Geometry. 2.8. Mathematical Formalization of Retrieval 2.9. Rag Pipeline: 2.10. Workforce Analytics Multi Agency Synergy. 2.11. Identity Mapping and Integration Agent. 12. Data Analysis Agent: Cognitive Architectures. 2. 13. Problems and Strategy Calibration. 2,14. The Privacy vs. Utility Trade-Off. 2.15. Optimal Reduction of the Hallucinations by Reranking. 2.16. Al Workforce Analytics that Construct Privateness. 2.17. Federated Learning and Differential privacy 2.18. Consequences of Automated Individual Decision-Making. 2.19. Hybrid Architectures: The Prodex Synergy... 2.20. Autonomous Cognitive vs. Workflow-Driven. 2.21. RAG Architecture on the Background of IT Services. 2.22. Development of Human Resource Management (1920-2025) 2.23. Four Phases in the Evolution of Human Resource Management 2.24. The Catalyst of the Covid-19: Remote Work & Digital Acceleration (2020-2022 )...... 2.25. The Present Landscape: The Hybrid Workforce and the Information Gap (2023-2025 2.26. The Rationality why a Multi-agent Framework is required. 2.27. Enhancing Interaction via intelligent Retrieval. 2.28. HR Systems Technology and Analytical Enablers. 2.29. Strategic Performance Management: Metric Evolution. 2.30. Information Infrastructure and Governance. 2.31. Chroma DB and Vector Storage. 2.32. RAG Ecosystem: Architecture and Review. 2.33. Foundations of Artificial Intelligence and NLP. 2.33.1 Gemini 1.5 Flash and Gemini 2.0 Flash-exp Model Architectures 2.33.2 The Productivity Data Sources: the DevOps Triumvirate. 2.33.3 Unified Activity Graph (UAG). 2.33.4 Holistic HR Analytics (HHRA): 2.34. Research Gap Identification. 2.35. Lack of Applied MAS-RAG Synergies in HR Analytics. Gap 1: Theoretical Abstraction and Enterprise Application. Gap 2: The Contextual Bridge Gap in Tech Sector HR. Gap 3: Unstructured data Privacy-Preserving Architectures. Gap 4: Dynamic vs. Static Knowledge Management. 2.36. 2.15. Summary of Contribution 3. Methodology (The PRODEX Framework). 3.1. Research Methodology: Design Science and Pragmatism.. 3.2. Prodex Multi- Agent System (MAS) Framework. 3.3. Data Integration Agent. 3.4. Decision Agent Formal Query Routing and Access Protocol. 3.5. Agent and RAG Pipeline Detail Embedding. 3.6. Data Analysis Agent. 3.7. HR Agent and Conversational Agent 3.8. Three Pillars of RAG Architecture in Prodex. 3.9. Agent Internals: The Black Box Revelation. 3.10. Integration Agent (The Spine )....... 3.11. Information Processing Agent (The Brain). 3.12. Productivity Analysis Agent (The Engine). 3.13. Long-term Data Modeling: Entity Relationship Diagram (ERD). 3.14. Techniques and Tools of Analysis: Stack Justification. 3.15. Safaricom Data Model and KPI. 3.16. System Evaluation and Metrics. 3.17. Provision of strategic technical justification and comparison analysis. 3.18. Multi-Agent Orchestration and ETL (Gemini vs. Alternatives) 3.19. Data Privacy and Governance Guidelines 3.20. Data Anonymization and Protection. 3.21. Bias Mitigation and Fair Processing 3.22. Algorithmic Transparency and Explainability. 3.23. Data Accuracy and Integrity 4. Data Analysis 4.1. Uncovering Information. 4.2. Working with the Hierarchical KPI Catalogue. 4.3. Slicing the Unified Activity Graph. 4.4. Findings Presentation. 4.5. How the Analysis is Surfaced 4.6. Insights. 4.7. Insight 1 A Hierarchical Catalogue Lets Each Audience See What They Need. 4.8. Insight 2 The Framework Refuses to Hallucinate. 4.9. Insight 3 Security and Utility Are Not in Conflict. 5. Conclusion 5.1. Problem Statement Recap. 5.2. Methodology Recap. 5.3. Key Insights 5.4. Implications. 5.5 User Interface Screenshots. 5.7. Future Work and Productisation Strategy 5.8. Evaluation Metrics 5.9. Evaluation Methodology. 5.10. User Profile and Organisational Directory 5.11. Task and Project Tracking 5.12. Resource Allocation and Reasoning 5.13. Document RAG and HR Policies 5.14. KPI Analytics and Performance Metrics 5.15. System Operations and Access Control 5.16. Role-Based Access Control Verification 5.17. System Guardrails, Ethics, and Hallucination Prevention 5.18. Reporting and Synthesis 5.19. Aggregate Scoring across Categories 5. Discussion. 7. References. Abstract This applied research tackles a structural problem in technology organizations: employee activity data is fragmented across SaaS platforms (GitHub, Jira, in-app messaging), producing semantic silos, reactive decisions, and an inability to measure intellectual productivity reliably. This project designs, implements, and evaluates Prodex, a privacy-preserving Multi-Agent System (MAS) framework based on Retrieval-Augmented Generation (RAG). Prodex converts fragmented digital exhaust into real-time, role-appropriate intelligence on workforce productivity. Its three core agents are an Orchestration Agent (Gemini 1.5 Flash) for routing and intent classification; an Information Processing Agent (gemini-embedding-001 plus ChromaDB) for semantic indexing: and a Productivity Analysis Agent (Gemini 2.0 Flash). Prodex fills the basic gaps in the current literature. Rather than relying on gameable proxies such as lines of code, the framework builds a Unified Activity Graph that links technical output (GitHub) to project management (Jira) and human factors (in-app sentiment), so high-dimensionality KPIs can be computed across four hierarchical tiers global, project, individual, and operational. A key contribution of this work is that a Security-First RAG Framework has been implemented. The Decision Agent that is built into the system offers advanced security, with granular Role-Based Access Control through metadata filtering before the semantic-search phase, ensuring that sensitive signals are mathematically inaccessible to unauthorized roles Validation metrics are action- and efficiency-oriented. Keywords: Multi-Agent Systems (MAS), Retrieval-Augmented Generation (RAG), Unified Human Resource Intelligence, Data Fragmentation, Holistic HR Analytics, Gemini 1.5 Flash, Agentic Orchestration.

Date Published:2026

The role of multimodal AI in building consumer trust in the authentication of luxury products

Afzal, Aiemah [22L-6458]

Submitted in fulfillment of the requirements for the degree of Bachelor of science in Business analytics to the Department of Management sciences. Table of Contents 1. Introduction.. 1.1. Background and Context... 1.2. Problem Statement 1.3. Research Gap 1.4. Research Questions. 1.5. Research Objectives. 1.6. Significance of the Study 2. Literature Review. 2.1. Introduction to Literature Review. 2.2. Relevant Theories Providing Guidelines for Research 2.2.1. Trust in Automation Theory 2.2.2. Algorithmic Transparency Theory 2.2.3. Error Management Theory 2.3. Conceptual Background: Luxury Authentication & Counterfeiting.. 2.4. Al and Counterfeit Detection (Technical Approaches) 2.5. Multimodal Al and Explainable Authentication. 2.6. Consumer Trust in Al and Automated Systems 2.7. Conceptual Framework & Hypothesis Development. 3. Methodology. 3.1. Research Design and Methodology Overview 3.2. Population, Sampling, and Participants 3.3. Data Collection Procedure 3.4. Stimuli Development 3.5. Measurement of Variables and Survey Instruments. 4. Data Collection 4.1. Dataset Construction for Technical Component. 4.2. Model Training and Evaluation Procedure 4.2.1. ResNet-50 (Image-Only Baseline) 4.2.2. ROBERTa (Text-Only Baseline) 4.2.3. CLIP VIT-B/32 (Primary Multimodal Model) 4.3. Survey Data Collection.. 5. Analysis and Results. 5.1. Descriptive Statistics and Data Overview. 5.2. Reliability Analysis. 5.3. Manipulation Check. 5.4. Cell Means by Condition. 5.5. Error Pattern Analysis. 5.5.1. ResNet-50 Error Patterns 5.5.2. ROBERTa Error Patterns. 5.5.3. CLIP Error Patterns. 5.5.4. Implications for Real World Deployment 5.6. Hypothesis Testing. 5.6.1. H1: Al Correctness and Consumer Trust 5.6.2. H2. Error Explanation and Perceived Transparency 5.6.3. Perceived Transparency and Consumer Trust 5.6.4. Mediation of Perceived Transparency 6. Discussion 6.1. Overview of Findings 6.2. Theoretical Implications 6.3. Practical Implications.. 6.4. Limitations. 6.5. Future Research Direction 7. Conclusion References. Appendix. Abstract With the boom of online luxury retailers and resellers, there is a growing need for authentication systems that consumers can trust. While there are technical advances in artificial intelligence (AI) that increase the accuracy of luxury good verification, less is known about consumer perception and trust of verification results produced by Al, especially when the result is incorrect or a statement of how Al works is included. In this study, the effect of multimodal Al on consumer trust responses towards luxury product authentication is explored, examining the relationship between model performance and consumer trust responses. The research introduces a multimodal approach that combines image-based deep learning and text analysis, exploring the impact of Al correctness and error explanations on trust and the potential of perceived transparency as an intermediate factor. The technical pipeline is divided into three models: ResNet-50 is the baseline model using images as input, ROBERTa is the baseline model using text as input, and CLIP (VIT-B/32) is the key multimodal model which is integrated by the input of both image and text. The models were trained and tested with 2,100 product image-text pairs across three luxury categories-handbags, watches and shoes-as authentic or inauthentic. The results showed that CLIP had an accuracy rate of 94.05% and an inauthentic recall rate of 82.5%, surpassing the accuracy (91.67%) and inauthentic recall (72.5%) of the ResNet-50 baseline. These findings highlighted the potential of multimodal fusion in counterfeit detection. In the experimental survey, the designers present the participants with four conditions: correct and incorrect authentication results by an Al, and with and without an explanation provided by the Al, in a 2x2 between-subjects design. The experiment involves a 2x2 between-subjects design with Al correctness (correct vs. incorrect) and explanation presence (with vs. without). The results are likely to reveal that accurate Al categorizations boost consumer trust and that providing explanations of the error boosts perceived transparency, leading to a positive effect on consumer trust. Perceived transparency is hypothesized to mediate the relationship between error explanation and consumer trust. Keywords: Multimodal Al, Luxury product authentication, Consumer trust, Explainable AI, Perceived transparency, Al correctness, CLIP, ResNet-50, RoBERTa, Online luxury resale.

Date Published:2026

Smart waste Sorting assistanat : An AI powered recycling solution for Pakistan

Edrees, Areeba [22L-6525]

Submitted in fulfillment of the requirements for the degree of Bachelor of science in Business analytics to the Department of Management Sciences. Abstract List of Tables List of Acronyms 1. Introduction 1.1. Problem Statement 1.2. Environmental, Social, Economic Consequences 1.2.1. Environmental Impacts 1.2.2. Economic Impacts 1.2.3. Public Health Impacts 1.3. Stakeholders 1.3.1. Government and Municipal Authorities 1.3.2. Businesses, Recycling Industries, and NGOs 1.3.3. Consumers and the General Public 1.4. Potential Solution: CNN-Backed Image Sorting Device 1.5. Impact of the Proposed Solution 1.5.1. Enhanced Operational Efficiency 1.5.2. Boosted Consumer Engagement 1.5.3. Improved CSR and Brand Image 1.6. Alignment with Sustainable Development Goals (SDGs) 1.6.1. SDG 11: Sustainable Cities and Communities 1.6.2. SDG 12: Responsible Consumption and Production 1.6.3. SDG 9: Industry, Innovation, and Infrastructure 1.7. Objectives 1.7.1. Primary Objective 1.7.2. Secondary Objectives 1.8. Identification of Gaps and Opportunities 2. Literature Review 2.1. The Global Waste Crisis and the Recycling Contamination Problem 2.2. The Promise of Artificial Intelligence in Waste Management 2.2.1. Evolution of Technical Approaches: From Classification to Complex Detection 2.3. Theoretical Frameworks: Bridging Al with Human Behavior and Systemic Goals 15 2.3.1. Behavioural Nudge Theory 2.3.2. The Circular Economy Framework 2.4. The Pakistan-Specific Context: A Critical Research Gap 2.5. Critical Analysis and Synthesized Research Gaps 2.5.1. Strengths 2.5.2. Weaknesses and Synthesized Research Gaps 2.6. Contradictions and Future Research Directions 3. Methodology 3.1. Data Sources 3.1.1. Primary Dataset 3.1.2. Secondary Dataset 3.2. Brand and Product Selection 3.3. Annotation Metadata 3.4. Data Pre-Processing 3.4.1. Data Set Merging & Label Standardization 3.4.2. Train, Validation & Test Split 3.4.3. Model Architecture: MobileNet V2 Transfer Learning 3.4.4. Two-Phase Model Training 3.5. Analytical Techniques 3.5.1. CNN for Classification 3.5.2. Clustering (Unsupervised) 3.5.3. Regression 3.5.4. Descriptive Statistics 3.6. Deployment 3.7. Python Libraries 3.8. Data Privacy & Ethics 4. Data Analysis 4.1. Overview of the Dataset and Descriptive Statistics 4.2. Data Quality and Integrity Checks 4.3. Training & Validation Performance 4.4. Evaluation of Test Set & Model Performance 4.5. Consumer Facing Considerations 5. Conclusion 6. References. The proposed project introduces a waste-sorting machine, based on Al, which is proposed and created as a hybrid image recognition model, which is meant to minimize consumer level mis-sorting, and enhance the accuracy of recycling in Pakistan. The device applies custom dataset based on the significant brands that are present in Pakistan both locally and internationally. Such data is also intertwined with real-time human generated information to ensure that the device is relevant to real life situations. The approach mainly combines convolutional neural networks (CNN) to classify materials, metadata analytics to understand context, and a cloud-based API layer to add new images to keep growing and developing the dataset. The device is implemented using a Google Colab based interface, where the user is requested to upload a picture and can be shown, in real time, the type of material they are dealing with and whether it can be recycled, as a solution to the consumer education gap and the unified nature of disposal guidelines. The results prove that a scalable and user-friendly Al tool that increases the accuracy of waste sorting, facilitates recycling by the municipalities, and promotes a more environmentally responsible behaviour is feasible. This work introduces a practical, technology-driven framework that strengthens recycling ecosystems in developing countries by merging machine learning. real-time data collection, and civic sustainability initiatives. Keywords: waste sorting, recycling technology, enn classification, sustainability analytics, smart waste management.

Date Published:2026

Paybot : A multilingual ai-driven chatbot for merchant support

Zahoor, Samra [22L-6459]

Submitted in fulfillment of the requirements for the degree of Bachelor of Business analytics to the Department of Management science. Abstract. Acknowledgments List of Tables List of Figures List of Acronyms 1. Introduction 1.1. Background and Context 1.2. Problem Statement 1.2.1. Objective 1.2.2. Interest 1.2.3. Proposed Solution 1.3. Identification of Gaps and Opportunities 1.3.1. Research Gaps 1.3.2. Opportunities for Further Research 1.3.3. Identified Gaps in Existing Literature 1.4. Approach and Methodology 2. Review of Previous Literature 2.1. Literature Review Summary 2.1.1. Natural Language Processing for Conversational Agents 2.1.2. Conversational Agents and Customer Support in Pakistan 2.1.3. Urdu and Roman Urdu in Natural Language Processing 2.1.4. Code-Mixing and Low-Resource Languages 2.1.5. Code-Mixing and Emotion/Sentiment in Roman Urdu and Urdu English 2.1.6. Punjabi Dialects, Saraiki, and Computational Resources 2.1.7. Urdu Banking Corpora and Domain-Specific Conversational Systems 2.1.8. Roman Urdu Linguistic Challenges 2.1.9. Intent Classification and Slot Filling in Dialogue Systems 2.1.10. Intent Classification and Spoken Language Understanding.24 2.1.11. Synthetic Corpora and Low-Resource NLP 2.1.12. Text-Based Customer Support Chatbots 2.1.13. Data Augmentation, Weak Supervision, and Synthetic Corpora 2.1.14. Evaluation Metrics in NLU Systems 2.1.15. Misclassification and Error Analysis in Code-Mixed, Low-Resource Settings 2.2. Summary of Key Studies 2.2.1. Major Findings 2.2.2. Methodologies 2.3. Implications for Your Analysis 2.3.1. Influence on Approach 3. Methodology 3.1. Proposed Analytic Techniques 3.1.1. Analytical Techniques 3.1.2. Justification 3.2. Benefits to the Consumer 3.3. Packages Required 3.3.1. List of Packages 3.4. Suppression of Messages and Warnings 3.4.1. Handling Messages 3.5. Purpose of Each Package 3.6. Data Preparation 3.6.1. Data Source 3.6.2. Purpose and Details 3.7. Data Privacy Guidelines 3.8. Data Explanation 3.9. Data Cleaning 4. Data Analysis 4.1. Preprocessing and Text Normalization 4.2. Feature Engineering and Embeddings 4.3. Model Selection and Training 4.4. Evaluation Metrics 4.5. Conversational Flow and Response Mapping 4.6. Analytical Dashboard Integration 5. Results 6. Conclusion 6.1. Mapping Problem Statement Recap 6.1.1. Restate Problem 6.2. Methodology Recap 6.2.1. Data and Methods 6.3. Key Insights 6.3.1. Findings 6.4. Implications 6.4.1. Impact 6.5. Limitations and Future Work 6.5.1. Limitations 6.5.2. Future Improvements 7. References Appendix A: Tables Appendix B: Case Study Appendix C: Teaching Note Appendix D: How To Note. Abstract The rapid surge of digital payments in Pakistan increased the need for streamlined and scalable merchant-support solutions among POS deployment firms, payment processors, and acquiring banks. Traditional call-based support models are still costly, time-limited, and unable to handle the high volume of repetitive merchant queries often expressed in Roman Urdu, Punjabi dialects, Urdu script, and English. This linguistic diversity and informality make automated support challenging due to the absence of standardized corpora and domain-specific natural language resources. This project presents PayBot, a multilingual, Al-driven, text-based chatbot designed to automate merchant support within Pakistan's payments ecosystem. The project creates a domain specific taxonomy of intents, a synthetic trilingual dataset, and a classification system that can comprehend seller complaints related to POS. The study compares the performance of NLP-based intent classification in low-resource, code-mixed environments using structured annotation intents, slots, dialect labeling, noise patterns, and misclassification reasons. This project shows that a proper taxonomy, natural language normalization and dialect specific data can help the system to provide a more precise answer and minimize the grey box in the merchandise questions. PayBot provides added value to academic and industrial research by providing a multilingual dataset focused on POS, a useful taxonomy of intents for serving merchants, and a roadmap for businesses to develop and maintain their own small help desks, especially in small market countries, without having to provide many operators. The direct applications of this study are for banks, POS sellers and companies that provide payment tools. They can use PayBot and maintain an open support line at all times. Keywords: support chatbot, intent classification, low-resource languages, conversational Al.

Date Published:2026

Predictive Analytics and equitable intervention planning for urdan heat vulnerability in Lahore

Ilyas, MinAhil [22L-6487]

Submitted in fulfillment of the requirements for the degree of Bachelor of science in Business analytics to the Department of Management Sciences. ACKNOWLEDGEMENT FOR WORK CONTRIBUTION Abstract List of Tables List of Figures 1. Introduction 1.1. Problem Statement 1.2. Identification of Gaps and Opportunities 1.3. Brief Discussion of Approach and Methodology 2. Review of Previous Literature 2.1. Summary of Key Studies 2.1.1. Urban Heat Island (UHI) Formation and Land Surface Temperature (LST) 2.1.2.Composite Vulnerability Indices 2.1.3. Rapid City Expansion and Heat Intensification 2.1.4.Green Urban Interventions 2.1.5.Machine Learning and Time Series Forecasting in Urban Climate 2.2. The Implications of Analysis 3. Methodology 3.1. Data Sources and Scope. 3.2. Data Privacy Guidelines 3.3. Data Explanation 3.4. Data Cleaning 3.5. Final Dataset 3.6. Variable Summary 3.6.1 Temporal and Spatial Identifiers 3.6.2.Land Surface Temperature Variables 3.6.3. Population Variables 3.6.4.Built-up Density Variables (m² per 100m pixel) 3.6.5. Socioeconomic Variables (tehsil level, assigned to UCs) 3.6.6.Composite Index Variables 3.6.7.HVI Category Distribution 3.7. Proposed Analytical Techniques 3.8. Packages Required. 4. Data Analysis 4.1. Temporal LST Analysis (2013-2024) 4.1.1.Annual Summer LST Trend 4.1.2.Seasonal LST Pattern by Tehsil 4.1.3.Spatial Heat Exposure Analysis 4.1.4.Urban Heat Island Intensity Map 4.1.5. Urban Expansion Map (1990-2020) 4.2. Heat Vulnerability Index Analysis 4.2.1.HVI Choropleth Map 4.2.2. Top 20 Vulnerability Breakdown 4.2.3. Tehsil-Level Average HVI 4.3. Predictive Modelling 4.3.1.Random Forest Model Spatial LST Prediction 4.3.2.ARIMA Forecast-2030 LST Projection 4.3.3.HVI Current vs 2030 Comparison 4.4. Intervention Simulation Analysis 4.5. Interactive Decision-Support Prototype 4.5.1. Panel 1 Hotspot Visualization 4.5.2.Panel 2 2030 Forecast 4.5.3.Panel 3 Vulnerability Explorer 4.5.4. Panel 4- Intervention Simulator 4.5.5. Panel 5- Priority Ranking Engine 4.6. Prototype Summary 5. Conclusion 5.1. Key Insights 5.2. Implications 5.3. Limitations and Future Work 5.4. Benefits to the Consumers 6. References 7. Appendix Appendix A: Dataset Summary and Sources Appendix B: Intervention Cooling Coefficients and Literature Sources Appendix C: Model Performance Summary Appendix D: ARIMA Forecast Results Appendix E: HVI Score Distribution Across All 151 Union Councils. With the incressong rapidity of urbanization and population growth, Lahore is getting afficted by Urban Host Island (UHI) which is putting vulnerable communities at risk due to high heat. This project builds an Urban Heat Vulnerability Prediction and Intervention Planning System for the 151 Union Councils (UCs) of Lahore based on 12 years (2013-2024) of satellite-derived Land Surface Temperature (LST) data, population data, built-up surface data, and socioeconomic indicators. A Heat Vulnerability Index (HVI) is created by integrating heat exposure factors and social vulnerability factors. Random Forest and ARIMA models are used to forecast future LST trends until 2010. An Intervention Simulator assesses the effectiveness and cost-effectiveness of heat mitigation measures, and a Priority Ranking Engine helps allocate resources equitably among UCs. The system is developed into a dashboard for urban planners using Streamlit. The findings show an average warming rate of +0.144°C per year, highlight 33 UCs as Very High Vulnerability areas, and demonstrate that the green park is the most cost-effective intervention, resulting in an average cooling of around -1.2°C for PKR 10 million invested. The framework illustrates the potential of using geospatial data and analytics with machine learning to assist with climate resilience planning in Lahore. Keywords: Urban heat island, heat vulnerability index, predictive analytics, geospatial analysis, intervention simulation, urban planning, Lahore, SDG 11, SDG 13.

Date Published:2026

Factor affecting tax revenue in pakistan

Arshad, Yousuf [22L-8447]

Submitted in fulfillment of the requirements for the degree of Bachelor of science in Business analytics to the Department of Management sciences. Contents Abstract. Acknowledgment List of Tables List of Figures. List of Acronyms... 1. Introduction. 1.1 Background and Context.... 1.2 Pakistan's Taxation Structure: An Overview 1.3 Retail Sector Challenges in Pakistan...... 1.4 The Documentation Crisis and FBR Data Landscape. 1.5 Geographic Taxation Disparities in Lahore. 1.6 Problem Statement.. 1.7 Research Objectives. 1.8 Research Questions.... 1.9 Significance of the Study 1.10 Dissertation Contributions 1.11 Scope and Limitations...... 2. Literature Review 2.1 Tax Compliance: Theoretical Foundations. 2.1.1 The Allingham Sandmo Model. 2.1.2 Behavioral and Sociological Extensions 2.1.3 Third Party Reporting and Information Asymmetry 2.2 Tax Gap in Developing Economies. 2.2.1 Conceptualizing the Tax Gap 2.2.2 Revenue Mobilization Evidence from South Asia... 2.2.3 The Role of Threshold Effects 2.3 Retail Sector Underreporting 2.3.1 Structural Determinants of Retail Tax Compliance 2.3.2 Business Size and Compliance Behavior 2.3.3 Wholesale Retail Classification Ambiguity 2.4 Informal Economy and Tax Evasion... 2.4.1 Measuring the Informal Economy..... 2.4.2 Informality and Geographic Concentration. 2.4.3 Pakistan Specific Informality Research 2.5 GIS and Geographic Analysis in Taxation. suo; 2.5.1 Spatial Approaches to The tax system......... 2.5.2 K Means Clustering in Fiscal Applications 2.5.3 Geographic Tax Intelligence in Developing Countries 2.6 FBR Reforms and Enforcement Challenges 2.6.1 Digitalization and Its Limits 2.6.2 The Trader Politics of Retail Taxation 2.6.3 Section 147 and Advance Tax Mechanisms..... 2.7 Regional Taxation Disparities: South Asian Context 2.7.1 Subnational Tax Variation.......... 2.7.2 Punjab Revenue Authority and Provincial Taxation. 2.7.3 Urban Rural Tax Disparities 2.8 Synthesis and Research Gaps 3. Research Methodology 3.1 Research Design and Philosophical Stance 3.2 Dataset Description and Source.... 27 28 29 30 30 31 32 33 33 33 34 35 38 38 39 3.3 The Two Tier Data Verification Framework 41 ....... 4 3.3.1 Tier 1- Explicit Lahore Matching.... 3.3.2 Tier 2 Implicit Area Based Matching (Lahore Gazetteer Dictionary 3.4 Sequential Data Cleaning and Financial Exclusions 4 3.4.1 Stage 1 Removal of Records with Missing Taxable Income... 3.4.2 Stage 2 Removal of Records with Missing Tax Chargeable... 4 3.4.3 Stage 3 Removal of Records with Missing Gross Revenue 3.4.4 Stage 4 Removal of Statistically Out of Bounds ETR Records. 3.4.5 Stage 5- Removal of Logically Inconsistent Profitability Ratios ....... 3.5 Feature Engineering 3.5.1 Effective Tax Rate (ETR).. 3.5.2 Profitability Ratio......... 3.5.3 Tax to Revenue Ratio... 3.5.4 Revenue Binning 3.6 Dummy Variable Construction 3.7 Geographic Clustering Methodology 3.7.1 Geographic Coordinate Extraction 3.7.2 K Means Algorithm ....... 3.7.3 Optimal Cluster Selection 3.8 Outlier Treatment 3.9 Research framework Summary. pue suor 4. Results and Discussion....... 4.2 Effective Tax Rate Distribution and Interpretation 4.1 Descriptive Statistics Overview 4.3 Profitability Ratio Analysis 4.4 Tax to Revenue Ratio Assessment. 4.5 Section 147 vs. Section 137: Comparative Analysis. 4.6 Geographic Cluster Comparison. 4.7 Gulberg vs. Anarkali: Revenue Leader vs. Density Leader 4.9 Compliance Behavior Analysis. 4.9.1 Size Compliance Gradient 4.9.2 The Unclassified Section Problem 4.9.3 Declared vs. Probable Income Gaps. 4.10 Policy Interpretation..... 5. Conclusion and Recommendations 5.1 Summary of Key Findings. 5.2 Policy Recommendations.. 5.2.1 Immediate Enforcement Adjustments..... 5.2.2 Geographic Enforcement Reallocation. 5.3 Lahore Retail Intelligence Unit: A Proposal. 5.3.1 Proposed Mandate and Functions. 5.3.2 Staffing and Resource Requirements 5.4 Implementation Feasibility 5.5 Limitations of the Study 5.6 Future Research Directions References Appendices. Appendix A: Raw Dataset Cleaning Pipeline - Detailed Record Count Summary Appendix B: Lahore Gazetteer Dictionary - Complete Keyword List......... Appendix C: Business Activity Classification Distribution. Appendix D: Tax Section Classification Distribution...... Appendix E: Revenue Bin Analysis - Complete Financial Statistics. Appendix F: Tax Concentration Analysis.... Appendix G: Geographic Zone Detailed Statistics Appendix H: K Means Cluster Model Comparison - Evaluation Criteria. Appendix I: Feature Engineering - Formula Reference and Interpretation. Abstract The retail sector plays a significant role in Pakistan's economy; however, concerns persist regarding the level of tax compliance and revenue contribution from retail businesses. Despite the implementation of tax reforms such as Point of Sale (POS) integration and documentation initiatives, many retailers continue to underreport income and maintain inadequate financial records. This study examines tax compliance patterns among retail businesses in Lahore using data obtained from the Federal Board of Revenue (FBR) for the year 2021. The initial dataset consisted of 29,296 records, which was cleaned and validated by removing incomplete entries and records containing inconsistent tax information. The final dataset comprised 13,506 retail businesses operating within Lahore. To evaluate tax performance, three financial indicators were calculated: Effective Tax Rate (ETR), Profitability Ratio, and Tax-to-Revenue Ratio. The findings reveal substantial variation in tax compliance across businesses. The average Effective Tax Rate was found to be 7.5%, while the median ETR was only 2.0%, indicating that a large proportion of businesses contribute relatively low amounts of tax. Collectively, the analyzed businesses generated total revenues of PKR 266.3 billion but paid only PKR 4.164 billion in taxes, representing approximately 1.56% of total revenue. Geographic analysis was conducted by grouping businesses into eight zones within Lahore to examine spatial patterns in tax compliance. The results indicate that business concentration does not necessarily correspond with higher tax contributions, as some densely populated commercial areas generated lower tax revenues than comparatively smaller business districts. Furthermore, a comparison of tax categories showed that businesses assessed under Section 147 (Advance Income Tax) exhibited a significantly higher average Effective Tax Rate of 20.5% compared to 8.2% for businesses assessed under Section 137 (Admitted Income Tax), suggesting a stronger level of compliance among firms subject to advance taxation. Based on these findings, the study recommends the establishment of a Lahore Retail Intelligence Unit, the strategic allocation of enforcement resources across zones, and the adoption of data-driven monitoring mechanisms to identify irregular declaration patterns. The research contributes to the ongoing discussion on tax reform in Pakistan by providing firm-level evidence on retail sector tax compliance and highlighting opportunities to improve revenue collection and regulatory effectiveness. Keywords: tax compliance, retail taxation, effective tax rate, Lahore retail sector.

Date Published:2026

Ai-based restaurant recommendation system

Shafi, Ali [22L-6522]

Submitted in fulfillment of the requirements for the degree of Bachelor of science in Business analytics to the Department of Management sciences. Table of ContentsAbstractACKNOWLEDGEMENTList of Acronyms 1. Introduction 1.1. Problem Statement 1.1.1. Objective 1.1.2. Interest 1.2. Identification of Gaps and Opportunities 1.2.1. Research Gaps 1.2.2. Market Gaps 1.2.3. Opportunities for Further Research 1.3. Brief Discussion about Approach and Methodology 2. Literature Review 2.1. Comprehensive Overview of Existing Research 2.2. Summary of Key Studies 2.2.1. Major Findings 2.2.2. Methodologies 2.3. Implications for Your Analysis 2.3.1. Influence on Approach3. Methodology 3.1. Data Sources, Type, and Scope 3.1.1. Google Ratings 3.1.2. Twitter/X Data 3.1.3. Bank Discount Data 3.2. Data Sampling, Preparation and Descriptive statistics 3.2.1. Sampling Method 3.2.2. Primary Cleaning and Pre-processing Data 3.2.3. Descriptive Statistics 3.3. Suggested Analytic Methodologies 3.3.1. Content-Based Filtering . Ranking Multi-Criteria Decision Making 3.3.3. Sentiment Analysis and Trending Score Computation 3.3.4. Correlation Analysis 3.4. Justification of Techniques 3.5. Benefits to the Consumer 3.6. Packages Required 3.7. Message and Warnings Suppression 3.8. Purpose of Each Package 3.9. Data Preparation 3.9.1. Data Sources 3.9.2. Data Integration 3.9.3. Data Explanation 3.9.4. Data Characteristics and Summary 3.9.5. Missing Values and Data Integrity 3.9.6. Data Cleaning 3.9.7. Final Dataset 3.9.8. Variable Summary 3.10. Data Privacy Guidelines 3.10.1. Anonymous and Data Security 3.10.2. Prevention of Biases in the Data Collection, Data analysis, and Data Interpretati 3.10.3. Transparent Methods and Algorithms 3.10.4. Ensuring Data Accuracy and Integrity 3.10.5. Explicit Consent Usage of Data 4. Data Analysis 4.1. Business Focused Insights 4.1.2. Discount Strategy Analysis 4.1.3. Social Media and Cuisine Performance 4.1.4. Market Distribution Analysis 4.2. Overall Business Insights 4.3. Customer Insights 4.3.1. Customer Preference Analysis 4.3.2. Impact of Social Media on Customer Preference 4.3.3. Value-Based Decision Making 4.3.4. System-Based Recommendations 4.4. Overall Customer Insights .Conclusion References. The Pakistan market is rapidly evolving towards digitization. Restaurant industry is one of them. However, there is a huge gap in digitization in restaurant industry which this project aims to address and resolve. There are multiple data sources where customers have to roam around to take decision for restaurant. Addressing this gap, this project develops a unified, AI-based restaurant recommendation system that integrates multi-source data into a single decision-support platform for the easiness for customers. There will be three aspects which the model will be integrating; how Google rating foresee quality, what impacts bank discounts have on restaurants and sentiment analysis though social media i.e. Twitter X. The study collects structured data from the Google Places API, unstructured social media data from Twitter/X, and promotional data scraped from official banking websites. The model will be based upon the CBF approach as it works on attributes, user preferences for cuisine, location, sentiment analysis, and discounts. For the final recommendation, a questionnaire will be held through which local customers' opinions will be taken. After all this, a MCDM approach and the Weighted Sum Model is applied on key attributes which will give consumers a final top 3 restaurants according to his decision. Our project will be a key initiative not just for helping consumers but to help businesses as well by simultaneously providing actionable insights of customer behaviour and promotional effectiveness. With this unified system, this project addresses a critical technological gap in Pakistan's restaurant industry and establishes a scalable foundation for future enhancements in broader consumer decision domains.Keywords: restaurant recommendation system, market fragmentation, CBF, MCDM, WSM.

Date Published:2026

Mindtrack : Mental health chatbot

Sohail, Jayish Bin [22L-6502]

Submitted in fulfillment of the requirements for the degree of Bachelor of Business analytics to the Department of Management science. Abstract Acknowledgements List of Figures List of Acronyms 1. Introduction 1.1 Theoretical Foundation 1.2 Sustainable Development Goals 1.3 Research Problem 1.4 Research Questions 1.5 Research Objectives 2. Literature Review 2.1 Global Mental Health Landscape 2.2 Digital Mental-Health Interventions 2.3 Artificial Intelligence in Mental Health 2.4 Mental health chatbots 2.5 Design and Limitations of Conversational Agents 2.6 User acceptance and TAM 2.7 Research Gaps 3. Methodology 3.1 Introduction and Conceptual Framework 3.2 System Architecture and Design 3.3 Frontend Development and User Experience 3.4 Backend Workflow and Automation using nån 3.5 Retrieval-Augmented Generation (RAG) Pipeline 3.6 Data Source, Scope, and Characteristics 3.7 Data Preparation and Cleaning 3.8 Descriptive Statistics and Data Understanding 3.9 Analytical Techniques and Justification 3.10 Integration of Cognitive Behavioral Therapy (CBT) 3.11 Benefits to Users and Stakeholders 3.12 Packages, Tools, and Technologies 3.13 Data Privacy and Ethical Considerations 4. Data Analysis 4.1 Exploration 4.2 User Interface Analysis 4.2 Findings Presentation 4.4 Insights 5. Conclusion 5.1 Problem Statement Recap 5.2 Methodology Recap 5.3 Key Insights 5.4 Implications 5.5 Limitations and Future Work 6. References. Abstract Mental health problems are gradually becoming a major global issue. Consequently, they need support and assistance, which should be easily reachable and technologized. Our project, therefore, aims to move the next step towards this rapidly changing topic by developing MindTrack, a conversational chatbot for mental health that is emotionally aware, and supportive by means of the contextually relevant conversation, to assist the users. Data collection was divided into two stages. In the first phase the team got hold of open-domain conversational datasets, which were later utilized to train a language model. The datasets were publicly available and could be accessed on various platforms such as Kaggle, Hugging Face, and GitHub. The second stage the team targeted emotionally annotated datasets in order to build a module that can accurately recognize the emotion in the input text. After the teams finished working on the separate datasets, they merged them into one complete dataset that was subjected to an extensive preprocessing pipeline including the steps of text cleaning, punctuation removal, tokenization, contraction expansion, lemmatization, and normalization. These methods enabled the team to get high-quality and noise-free texts, which are suitable for modeling applications to be carried out downstream. The results obtained from the data preparation and EDA stages set up the framework for the next steps of the MindTrack system, which include deep learning which is based emotion classification and chatbot answer generation. Keywords: Mental Health Chatbot, Emotion Recognition, Natural Language Processing (NLP), Deep Learning, Machine Learning, Sentiment Analysis, Conversational AI.

Date Published:2026

Automated coronary angiography segmentation and stenosis detection : A comparative study of resunet and attention U-Net using a data inversion methodology on the arcade dataset

Abdullah, Muhammad [22L-6501]

Submitted in fulfillment of the requirements for the degree of Bachelor of science in Business analytics to the Department of Management Sciences. Abstract Acknowledgements Table of Contents List of Tables List of Acronyms 1. Introduction 1.1 Overview and Contextual Background 1.2 Problem Statement 1.2.1 Objective 1.2.2 Interest and Relevance 1.3 Research Context and Contributions 1.3.1 Research Gaps Addressed 1.3.2 Thesis Contributions 1.4 Discussion of Approach and Methodology 1.4.1 The Dataset: ARCADE Challenge 1.4.2 Methodology Strategy: Data Inversion Pipeline 1.4.3 Comparative Architecture Strategy 1.5 The structure of the thesis 2. Review of Previous Literature 2.1 Introduction 2.2 The Diagnostic Challenge: The ARCADE Study Context 2.2.1 The Crisis of Subjectivity 2.2.2 Structure of the Dataset and the Pathology Bias 2.3 Evolution of Segmentation Architectures 2.3.1 From U-Net to ResUNet 2.3.2 U-Net++ and Nested Skip Connections...Error! Bookmark not defined 2.3.3 Attention Mechanisms in Medical Image Segmentation........... Error! Bookmark not defined 2.4 Vesselness as a Guiding Prior 2.5 Identification of Gaps and Positioning 2.5.1 Research Gaps 2.5.2 Positioning of This Work 2.6 Summary 3. Methodology 3.1 Introduction and Methodological Framework 3.2 Dataset Acquisition and Characterization 3.2.1 Data Specifications 3.2.2 The Class Imbalance Challenge 3.3 The Data Inversion Pipeline 3.3.1 Phase I: Preprocessing and Artifact Removal 3.3.2 Phase II: Full-Width Vessel Detection 3.3.3 Phase III: Geometric Subtraction and Label Generation 3.4 Model Architecture: ResUNet 3.4.1 Architectural Design 3.5 Model Architecture: Attention U-Net (SCSE Variant) 3.5.1 Architectural Design 3.6 Training Strategy 3.6.1 Controlled Experimental Design 3.6.2 Hybrid Loss Function 3.6.3 Data Augmentation and Sampling Strategy 3.7 Evaluation Framework 3.7.1 Primary Metric: Dice Score 3.7.2 Qualitative Evaluation 4. Results and Discussion 4.1 Introduction 4.2 Dataset Validation: The Scale of Contextual Enrichment 4.2.1 Class Distribution After Inversion 4.3 Training Dynamics 4.3.1 ResUNet Training Progression 4.3.2 Attention U-Net Training Progression 4.4 Quantitative Comparison 4.4.1 Per-Class Dice Scores 4.4.2 Analysis of Results 4.5 Discussion 4.5.1 Interpretation of Results 4.5.2 Contextualizing the Dice Scores 5. Conclusion 5.1 Summary of Contributions 5.2 Implications 5.2.1 Clinical Implications 5.2.2 Methodological Implications 5.3 Limitations 6. Recommendations for Future Work 6.1 Short-Term Improvements 6.2 Long-Term Research Extensions 6.3 Toward Automated Virtual Fractional Flow Reserve References. Coronary Artery Disease (CAD) remains the leading global cause of mortality, and its diagnosis through X-ray Coronary Angiography (XCA) continues to rely heavily on subjective visual interpretation. This project addresses the challenge of inconsistent stenosis assessment by developing and comparatively evaluating two deep learning architectures for automated semantic segmentation of coronary angiograms. A critical limitation of the ARCADE dataset-which provides stenosis annotations but no labels for healthy vessels was overcome through a novel Weakly Supervised Data Inversion Pipeline. This pipeline employs Frangi Vesselness Filters, adaptive thresholding, and morphological operations to mathematically generate full-width healthy vessel labels, converting the dataset into a three-class semantic segmentation problem (Background, Healthy Vessel, Stenosis). Two architectures were trained under identical experimental conditions to isolate the effect of decoder-side attention: a ResUNet (U-Net++ with ResNet-34 encoder) and an Attention U-Net incorporating Spatial and Channel Squeeze-and-Excitation (SCSE) blocks. Both models were trained using a hybrid loss function combining Cross-Entropy, Tversky, and Focal Loss components, with weighted random sampling to address severe class imbalance (stenosis constituting less than one percent of image pixels). The ResUNet achieved a combined Dice score of 0.6210 (Vessel: 0.6561, Stenosis: 0.5976), while the Attention U-Net achieved 0.6194 (Vessel: 0.6471, Stenosis: 0.6009). Results confirm that each architecture excels where its design predicts: ResUNet produces smoother, more continuous vessel boundaries through nested skip connections, whereas Attention U-Net marginally improves stenosis localization through background noise suppression. This work demonstrates a practical, reproducible path toward automated CAD assessment and establishes a controlled architectural comparison framework for medical image segmentation under weak supervision. Keywords: coronary artery disease, semantic segmentation, ResUNet, Attention U-Net, Dice score, data inversion, ARCADE dataset, weakly supervised learning.

Date Published:2026

Hippa compliance Service

Kashif, Moaiz [21L-6156]

Submitted in fulfillment of the requirements for the degree of Bachelor of Business analytics to the Department of Management science. Table of Contents ABSTRACT ACKNOWLEDGEMENT. EXECUTIVE SUMMARY INTRODUCTION. Summary of Competitive Strengths SERVICE PORTFOLIO.. SECURITY RISK ASSESSMENT POLICIES AND PROCEDURES DEVELOPMENT HIPAA COMPLIANCE INTRODUCTION TO POLICIES AND PROCEDURES. THE POLICY DEVELOPMENT AND THE STRATEGIC ROLE IN COMPLIANCE GOVERNANCE... AN OVERVIEW OF THE ADMINISTRATIVE SAFEGUARDS OF HIPAA. OUR FIRM'S METHODOLOGY FOR POLICY AND PROCEDURE DEVELOPMENT INDIVIDUALIZATION AND FLEXIBILITY OF POLICY INSTANCES.. STRIKING A BALANCE BETWEEN REGULATORY RIGOR AND OPERATIONAL FLEXIBILITY. BASIC CATEGORIES OF HIPAA-COMPLIANT POLICIES. POLICY DOCUMENTATION LIFE CYCLE: CREATION TO REVISION. PROBLEMS IN IMPLEMENTATION AND BARRIER SOLUTIONS. ASSURING STAFF ENGAGEMENT AND POLICY ADOPTION. BRINGING REGULATORY FRAMEWORKS AND BEST PRACTICES INTO ALIGNMENT. VERSION MANAGEMENT AND DOCUMENT CONTROL. POLICY ENFORCEMENT AND ACCOUNTABILITY STRUCTURES ON-GOING THE IMPROVEMENT BY USE OF COMPLIANCE MATURITY MODELS. SOME CRITICAL HIPAA POLICY AREAS. Access Control Policy:. Information Retention and Destruction Policy:.. Incident Response and Breach Notification Policy:. Vendor and Third-Party Management Policy:.. Mobile Device Policy and Remote Access Policy:. INCORPORATION OF POLICIES INTO ORGANIZATION CULTURE POLICY MANAGEMENT, ENABLING AND AUTOMATING TECHNOLOGY.. REINFORCEMENT MECHANISM OF TRAINING. AUDIT READINESS AND DOCUMENTATION TRACEABILITY. ENTERPRISE GOVERNANCE AND RISK MANAGEMENT INTEGRATION. LONG-TERM STRATEGIC BENEFITS OF POLICY DEVELOPMENT INVESTOR CONFIDENCE AND CREDIBILITY OF THE MARKET THE CHANGING TREND IN THE INDUSTRY AND MODERNIZATION OF THE POLICY. ETHICAL PRINCIPLES OF POLICY FORMULATION.... THE FUTURE OF HEALTHCARE COMPLIANCE POLICY CREATION.. SUMMARY: THE POLICIES, THE PILLARS OF THE SUSTAINABLE COMPLIANCE.. VAPT FOR APPLICATION. APPLICATION SECURITY AND VAPT INTRODUCTION, THE BROKEN LANDSCAPE OF THREATS TO HEALTHCARE APPLICATIONS PURPOSE AND OBJECTIVES OF APPLICATION VAPT OUR FIRM'S VAPT METHODOLOGY. Planning and Scoping :.......... Discovery and Enumeration: Vulnerability Analysis: .......... Exploitation and Validation:.. Reporting and Remediation Advice ............ DIFFERENTIATING BETWEEN VULNERABILITY ASSESSMENT AND PENETRATION TESTING....... TYPICAL HEALTHCARE VULNERABILITIES IN APPLICATIONS Injection Attacks (SQL, Command, and LDAP Injection): Cross-Site Scripting (XSS):.. Broken Authentication and Session Management:...... Insecure Direct Object References (IDOR):. Security Misconfigurations .......... Sensitive Data Exposure: Inadequate Logging and Surveillance: LESSONS LEARNED AND CASE STUDIES IN HEALTH CARE. Case Study 1: Access to Unsecured Patient Portal..... Case Study 2: Incorrectly set up Cloud Application and Data Exposure. Case Study 3: Telehealth startup with Weak Authentication Controls. IMPORTANT LESSONS OF THE CASE STUDIES.. CONTINUOUS TESTING AND THE DEVSECOPS MODEL DETAILED FRAMEWORK AND REPORTING OF THE RESULTS... SETTING AND IMPLEMENTING REMEDIATION MEASURES INTEGRATION WITH COMPLIANCE, RISK AND GOVERNANCE SYSTEMS. THE BUSINESS AND INVESTOR CASE OF ALIGNING TO REGULAR VAPT ENGAGEMENTS. Building Brand and Reputation:... Minimizing Financial Exposure:. Meaningful Reinforcement of Investor Confidence: HELPING TO PLUG THE GAP: CYBER INSURANCE ELIGIBILITY MAKING COMPLIANCE STRATEGIC GROWTH.. THE FUTURE OF APPLICATION SECURITY TESTING AND THE NEW TECHNOLOGIES DEVELOPING A CULTURE OF ETHICAL HACKING THE INVESTOR VIEW: CYBERSECURITY AS A VALUE MULTIPLIER VAPT FOR CLOUD/NETWORK. CLOUD AND NETWORK VULNERABILITY ASSESSMENT INTRODUCTION GOALS OF CLOUD AND NETWORK VAPT THE CLOUD AND NETWORK VAPT METHODOLOGY IN OUR FIRM.. SCOPING AND ASSET DISCOVERY VULNERABILITY TESTING AND SYSTEM SECURITY REVIEW Exploitation Simulation and Penetration Testing Privilege Testing and Privilege Escalation Testing Executive Presentation and Reporting. CLOUD INFRASTRUCTURE-SPECIFIC ASSESSMENT AREAS. TESTING OF NETWORK LAYER AND RESILIENCE OF ARCHITECTURE....... Case Study 1: PHI Exposure by a misconfigured Cloud Storage......... Case Study 2: Ransomware attack through the unpatched network vulnerability. Case: 3 Insider Misuse and Network Segmentation Failure....... HIPAA AND CLOUD/NETWORK COMPLIANCE MAPPING OUR FIRM'S REMEDIATION FRAMEWORK FOR CLOUD AND NETWORK VAPT Immediate Mitigation. Root Cause Analysis: Verification and Retesting:. Strategic Improvement: LESSONS LEARNED AND STRATEGY IMPLICATIONS EXTENSIVE REPORTING AND RISK COMMUNICATION STRUCTURE. Technical Vulnerability Report: Compliance Mapping Report: Executive Risk Summary: .......... Graphic Dashboards and Trend Analysis: Automation and Continuous Security Monitoring Continuous Vulnerability Scanning: Cloud Security Posture Management (CSPM):... Security Information and event management (SIEM) Integration: Automation of Reporting and Alerts:......... INTRODUCTION TO CLOUD GOVERNANCE AND COMPLIANCE ARCHITECTURE. DEVELOPING A CLOUD SECURITY MATURITY MODEL.. THE INVESTOR RELEVANCE OF CLOUD AND NETWORK VAPT INTERNATIONAL REGULATORY DEVELOPMENT AND INTERNATIONAL IMPLICATIONS.. ORGANIZATIONAL AND STAKEHOLDER STRATEGIC BENEFITS........... FUTURE OF CLOUD AND NETWORK SECURITY: IN THE DIRECTION OF PREEMPTIVE, AUTONOMOUS DEFENSE.. CONCLUSION: SECURING A COMPLIANT, TRUSTED AND FUTURE INFORMATION SECURITY OFFICER ENABLEMENT INTRODUCTION TO INFORMATION SECURITY OFFICER ENABLEMENT THE STRATEGIC ROLE OF THE INFORMATION SECURITY OFFICER PROBLEMS IN DEVELOPING GOOD SECURITY LEADERSHIP. Talent Shortage:. Ambiguous Role Definition:. Poor Governance Structures. Regulatory Pressure: Cultural Resistance:. THE ENABLEMENT FRAMEWORK OF OUR FIRM WITH RESPECT TO INFORMATION SECURITY OFFICER Phase 1: Evaluation and Gap Analysis Phase 2: Role definition, Role Governance.. Phase 3: Building Capacity and Mentoring. Phase 4: Implementation and Integration Phase 5: Ongoing Support and Performance Appraisal. BASIC INFORMATION SECURITY OFFICER RESPONSIBILITIES INTEGRATION WITH HIPAA AND REGULATORY FRAMEWORKS DEVELOPING SECURITY LEADERSHIP COMPETENCY. Case example 1: The ISO Enablement of a Mid-Sized Hospital Network. Case Example 2: ISO Enablement in a Health Insurance Firm INCIDENT RESPONSE TEAM ENABLEMENT INTRODUCTION TO INCIDENT RESPONSE IN HEALTHCARE ORGANIZATIONS. INCIDENTS RESPONSE TEAMS: STRATEGIC SIGNIFICANCE. Strategic Dimension: Operational Dimension: Cultural Dimension:...... Incident Response Preparedness is faced with common challenges. Absence of Formulated Response Structures:. Lack of Cross-Functional Integration: Poor Forensic Capabilities: Insufficient Training and Simulation Training: Phase 1: Reading and Preparation Assessment. Phase 2: Team Organizing and Functionalization... Phase 3: Incident Response Policies and Playbooks Development. Phase 4: Simulation and Tabletop Exercises. Phase 5: Onward Improvement and Integration of Automation... EFFECTS OF EFFECTIVE IRT ENABLEMENT.... 1. Operational Benefits:... 2. Strategic Benefits: ............... HIPAA SECURITY RULE INCORPORATION. COMPLIANCE WITH BREACH NOTIFICATION RULE.. INTERRELATIONSHIP WITH OTHER REGULATORY FRAMEWORKS INCIDENT LIFECYCLE MANAGEMENT. Preparation Detection and Analysis. Containment. Eradication Recovery. Lessons Learned STAKEHOLDER CO-ORDINATION AND STRATEGY OF COMMUNICATION. Internal Communication.... Executive and Board Briefings. Outbound Communication and PR.. Coordination between the regulators and partners. Learning after Incidents and Organizational Maturity. Root Cause Analysis (RCA)..... Policy and Process Updates:. Metrics Tracking: Cultural Reinforcement:. Automation and Security Orchestration Integration. Robotization of Monotonous work. Major Incident Co-ordination ........... Intelligence on threat. Al-Assisted Analysis. Key Performance Indicators (KPIs) Feedback and Improvement Loops Cross-Functional the Cooperation and Organizational Synergy. IT Operations Integration.. Cooperation with Compliance and Legal Teams.. HR and Employee Engagement. Executive Supervision and Board Reporting... Investor and Business Relevance... The Ethical Governance and Sustainability of Incident Response. Retention and Development of Skills in the long- term............. Information Stewardship and Ethics The documentation and Institutional Learning. Succession Planning Assimilation into Greater Organization Strategy. Governance Alignment. BUSINESS CONTINUITY PLANNING (BCP). INTEGRATION WITH COMPLIANCE AND QUALITY MANAGEMENT IMPROVING STRATEGIC DECISION-MAKING. THE INVESTOR VIEW: CYBER RESILIENCE: A BUSINESS ASSET.. CONCLUSION: RESPONSE TO RESILIENCE. STAFF/EMPLOYEE TRAINING INTRODUCTION: THE HUMAN FACTOR IN INFORMATION SECURITY STRATEGIC VALUE OF WORKFORCE SECURITY TRAINING INCREASED REGULATORY COMPLIANCE.. ENHANCED CYBER RESILIENCE CULTURAL TRANSFORMATION ISSUES WITH EMPLOYEE SECURITY TRAINING Training Fatigue Lack of Customization Insufficient Reinforcement. Leadership Apathy............. Lack of Measurable Outcomes Employee Enablement Framework of our Firm.... Phase 1: Needs Assessment Phase 2: Curriculum Design.. Phase 3: Implementation and Delivery. Phase 4: Behavioral Conditioning and Reinforcement.. Phase 5: Measurement and Continuous Improvement. THE PSYCHOLOGY OF SECURITY AWARENESS RISK PERCEPTION AND COGNITIVE CONDITIONING THE USE OF REPETITION AND REINFORCEMENT THE PSYCHOLOGY OF MOTIVATION. BREAKING THE HOLD OF COGNITIVE BIASES. REINFORCEMENT OF BEHAVIOR IN AN ORGANIZATION. Leadership Modeling....... Positive Recognition and Reconstruction.. Assimilation into HR Policies Constant Environmental Verbal Cues..... Training Technologies and Training Models. Experiential Learning.... Role-Based Training ...... Gamification Simulation-Based Learning Hybrid Learning Environment The Emotional Intelligence in Cybersecurity Training. Managing Fear and Stress Building Empathy Promoting Mindfulness. THE LONG-TERM VISION: THE AWARENESS TO EMPOWERMENT Qualitative Indicators...... Constant Reduction by means of feedback and analytics.. Adaptive Learning Pathways..... Detailed Annuals: Content Refresh Cycles.... Feedback-Driven Refinement...... Benchmarking and Peer Comparison. ENGAGEMENT AND ORGANIZATION INTEGRATION LEADERSHIP Executive Participation..... Security Steering Committes Communication from the Top. Relevance of Investors and Stakeholders ..... Human Risk as an objective Factor. Compliance Confidence. Empowering Reputation and Patient Trust. Institutionalizing a Culture of Lifelong Learning THE WORKFORCE SECURITY AWARENESS OF SUSTAINABILITY... Micro-Sustainability Model. Knowledge Retention and Institutional Memory. Ecological and Material Eco-efficiency. Mentorship and Leadership Continuity Adaptive Refresh and Review Cycles.... ESG AND CORPORATE RESPONSIBILITY ALIGNMENT Environmental (E) ........ Social (S). Governance (G)..... CONCLUSION: HUMAN FIREWALL EMPOWERMENT BREACH SUPPORT SERVICE INTRODUCTION: REACTING TO THE NEW CYBER BREACH REALITY. 19 19 BREACH SUPPORT: THE STRATEGIC IMPORTANCE 19 PROTECTION OF PATIENT TRUST.... HIPAA regulatory Compliance. 19 Financial Continuity and Operational Continuity.. Investor and Stakeholder Assurance.... Systemic Breach Response Problems. Delayed Detection. Disorganized Coordination.. Complexity in Law and Regulation.... Emotional and Cultural Fallout. Poor Public Relations Management ........ Breach Support Service Framework of our Firm Phase 1 Rapid Detection and Triage Phase 2: Legal Analysis and Impact Assessment. Phase 3: Stakeholder and Patient Communication... Phase 4: System Recovery and reinforcing Phase 5: Post-Breach Analysis and Reporting. Results and Long-Term Effects.. Regulatory Congruence and Lawful Requirements in Reaction to a Breach HIPAA BREACH NOTIFICATION RULE COMPLIANCE. HITECH Act and State Privacy Laws. International and Cross-Border Issues. The Ethics of Disclosure............ Internal Communication and Co-ordination Media and Public Relations.. Root Cause Analysis and Technical Forensics Evidence Preservation. Event Reconstruction and Data Analysis. Threat Intelligence and Malware.. Vulnerability and Control Assessment. Forensic Reporting and Legal Integration.. Compliance and Documentation Methodology of our Firm. Breach Response Reporting Framework. Audit Readiness........ Compliance Systems Integration... Continuous Legal Oversight The Merit of an Organized Response to a Breach.. Institutional Resilience and Organization Learning From Incident to Insight. Development of Knowledge Repository. Policy and Training Programs Associations Leadership Involvement in Learning Performance Measurement and Continuous Improvement. Benchmarking and Industry Comparison. Continuous Optimization. Relevance of investors and Stakeholders.... Investor Confidence by Governance..... Minimization in Financial and Insurance Risk Competitive Advantage as Regulatory Transparency.. Strategic Stability and Brand Worthiness.. CONCLUSION: REACTIVE RESPONSE TO STRATEGIC READINESS.. ESG FIT AND RESPONSIBLE BREACH MANAGEMENT ETHICS Environmental (E), Digital Recovery of Sustainability.... Social (S), Saving People and Healing Trust.. Governance (G), Accountability and Oversight Recovery of Patient and Public Trust Open Community Firms.... Long-term Communication Strategy. Strengthening Ethical Culture The partnership with Patient Advocacy Groups. Cyber Resilience and Long-Term Sustainability. Technology Renewal Long Term Economic Stability. Business Continuity Planning integration. Ethical Responsibility and Prospective. Reflections on Crisis into Competence. COMPLIANCE AUDIT SERVICE. BACKGROUND: THE FOUNDATION OF TRUST AND ACCOUNTABILITY. THE STRATEGIC VALUE OF COMPLIANCE AUDITING Safeguarding Patient Data..... Increasing the Organizational Reputation. Minimizing Financial and Regulatory Risk. Favoring Strategic Decision-Making....... Building on Investor Confidence. Precision.... Independence. Integration.. Value Creation.... The Compliance Audit Lifecycle. Planning and Scoping.. Identification and Mapping of Control.. Collection and Testing of Evidence.. Gap Analysis and Risk Evaluation... Report and Recommendation Remediation Support and Follow-Up. Advantages of Compliance Audit Service by Our Firm.. The Drift toward Risk-Based Auditing. Understanding Risk Context... Prioritization and Weighting. CONTINUOUS RISK MONITORING Cloud-Based Auditing Management Systems. The Audit Ethics....... The Confidentiality Worthy of respect. HEALTHCARE GOVERNANCE COMPLIANCE AS A PILLAR OF HEALTHCARE GOVERNANCE. RISK, ETHICS AND GOVERNANCE...... Measuring Performance and Maturity of Compliance. 1. Setting up Compliance Key Performance Indicators (KPIs). 2. Compliance Maturity Model......... AUDITING REPORTING AND TRANSPARENCY SYSTEMS...... Detailed Audit Report....... Visualization and Analytics THIS CRITICAL TRANSPARENCY IS THE GAP BETWEEN AUDIT AND STRATEGIC GOVERNANCE. Verification and Assurance of the Third Party. CONSTANT IMPROVEMENT AND INCLUSION OF FEEDBACK Compliance is not a condition, rather it is an ongoing evolution process. Post-Audit Review Workshops Improvement of Feedback-Based Methodology. REMEDIATION TRACKING AND VERIFICATION. Constant improvement needs an insight into progress. Maturity Assessment of Compliance on a yearly basis. Investor and Stakeholder Compliance Maturity The Governance Indicator of Compliance. Inclusion in the ESG Reporting. Risk Financing and Insurance... Market Differentiation and Public Confidence..... CONCLUSION: THE COMPLIANCE AS STRATEGIC CAPITAL MEASUREMENT. Digital Compliance Auditing Transformation....... Cloud-Native Compliance Systems.... Audit Integrity with the help of blockchain. Online Cooperation and Availability. Sustainability of Compliance Programs in the Long Run Developing Continuous Auditing as an Institution..... Knowledge Retention in Compliance..... Integration with Strategic Planning. Cross-Functional Collaboration. Scalability as Future Proofing... Summary: Compliance as a Push towards Sustainable Excellence. PRICING MODEL AND SKILL LEVEL STRATEGY The Tiered Skill-Level Framework..... Value Alignment with Market Realities Economic Efficiency and Transparency of Costs. The Tiered Pricing Strategic Rationales......... Flexibility on Market and Project Conditions Investor and Market Perspective OPERATIONAL TIME ESTIMATES Time Allocation Philosophy. Service-Wise Time Distribution HOUR ESTIMATION RATIONALE.. Core Task Complexity:. Composition of Team and Expertise Mix:... Depth of Compliance and Documentation:.. Automation of Operational Processes and Time Management. Scalability between Organization Types Efficiency Metrics and Performance Tracking. The Role of Technology in Time Management... INVESTOR AND FINANCIAL RELEVANCE FINANCIAL AND OPERATIONAL ANALYSIS REVENUE MODEL: A HYBRID BETWEEN PREDICTABILITY AND GROWTH. Recurring Services: Project-Based Services:... Margins in profits and Cost Structure. Cost Allocation by Skill Level.... Operating Expense Management. Resource Overlap Optimization .... Dynamic Project Allocation. Robotization and Electronics... Service Cost Benchmarking FINANCIAL SUSTAINABILITY AND CASH FLOW STABILITY SCALABILITY AND MARKET FIT LEARNING ABOUT MARKET DYNAMICS EXPANSION SERVICE ARCHITECTURE Technological operational Scalability Market Fit and Competitive Advantage. Flexibility to Regulatory Evolution............ Growth Potential and Expansion Pathways. Investor Perspective: Scalability as a Value Multiplier. INVESTMENT OUTLOOK........ Uniting Compliance, Innovation, and Trust.. The Intersection of Technology and Government. Operational Strength and Financial Prosperity. Social responsibility and ethical Foundation. FUTURE STRATEGIC VISION Constructing Future of Confidential Healthcare. SCREENSHOTS REFERENCES. Abstract The project creates an HIPAA Compliance Service model structure and its comprehensive form to assist healthcare organizations to protect Protected Health Information (PHI) and regulatory requirements and improve their overall cybersecurity posture. As the digitalization of healthcare continues to rise (electronic health records, cloud systems, and interconnected healthcare systems), healthcare organizations are exposed to security breaches, operational vulnerabilities, and regulatory fines. In reaction to these issues, the suggested framework incorporates the fundamental elements of compliance, such as Security Risk Assessments (SRA), policy and procedure formulation, Vulnerability Assessment and Penetration Testing (VAPT), compliance auditing, breach response coordination, and Information Security Officer (ISO) enablement. The model conforms to the standards of other relevant programs including HIPAA, HITECH, and NIST where both administrative and technical protection mechanisms are taken care of in a systematic and evidence-based approach. It focuses on maturity of governance, policy lifecycle management, on-going monitoring and how important the workforce awareness is in maintaining long term compliance. Moreover, the project will present a financial viable service architecture with a hierarchical skill-level pricing approach and mixed revenue system, which can be economically affordable to small practices and can be extended to large healthcare organizations. This project reveals how an integrated compliance service will enhance an organizations resiliency, minimize both legal and financial liabilities, and build trust with patients by balancing regulatory rigor, operational viability and technological flexibility. The ensuing framework is a realistic and forward-thinking solution that can be used to meet the dynamic cybersecurity and compliance requirements of the healthcare sector.

Date Published:2026

From button phones to smartphones the role of AI in consumer shift

Shahid, Nazish [22L-6473]

Submitted in fulfillment of the requirements for the degree of Bachelor of Business analytics to the Department of Management science. Table of Contents Abstract Acknowledgement 1. Introduction 1.1. Background of Study 1.2. Research Problem 1.3. Research Question 1.4. Research Objectives 1.5. Research Gap 1.6. Scope of the Study 1.7. Significance of Study 2. Literature Review 2.1. Al Features 2.1.1. Role of Al Features in User Convenience 2.1.2. Al Features and Personalization in Smartphones, 2.1.3. Al Features and Automation in Daily Smartphone Use, 2.1.4. Al Features and Perceived Usefulness 2.2. Psychological and Social Processes 2.2.1. Social Identity and Smartphone Adoption, 2.2.2. Psychological Attachment to Smartphones 2.2.3. Social Influence and Peer Pressure 2.2.4. Fear of Missing Out and Smartphone Adoption 2.3. AI Influence (Mediator) 2.3.1. Al Influence as a Behavioral Mechanism 2.3.2. Al Influence and User Trust 2.3.3. Al Influence and User Dependency 2.3.4. AI Influence as a Link Between Features and Adoption 2.4. Age as a Moderating Factor 2.4.1. Generational Differences in Al-Based Smartphone Adoption 2.4.2. Age and Digital Literacy 2.4.3. Age and Technology Anxiety 2.5. Smartphone Adoption Rate (Dependent Variable) 2.5.1. Smartphone Adoption as a Behavioral Shift 2.5.2. Decline of Button Phones 2.5.3. Smartphone Adoption and Consumer Satisfaction 2.6. Integration of Research Variables 2.6.1. Theoretical Foundation of the Study 2.6.2. Technology Acceptance Model and Smartphone Adoption 2.6.3. Diffusion of Innovation Theory and Smartphone Adoption 2.6.4. Conceptual Framework Explanation 2.7. Development of Hypotheses 2.8. Integration of Variables and Research Implications 2.9. Summary 3. Research Methodology 3.1. Research Design 3.1.1. Justification for Quantitative Research Design, 3.2. Research Type - Secondary Research 3.3. Population 3.4. Sampling Technique 3.4.1. Justification for Purposive Sampling, 3.5. Sample Size 3.5.1. Sample Size Adequacy for PLS-SEM 3.6. Demographics (Planned Variables) 3.7. Data Collection Method 3.8. Questionnaire Design 3.8.1. Measurement Scale 3.9. Variables 3.10. Data Preparation and Cleaning 3.10.1. Data Preparation and Cleaning 3.10.2. Handling of Missing Values 3.11. Data Analysis Technique, 3.11.1. Reason for Using SmartPLS 4 3.11.2. PLS-SEM Analysis Procedure 3.12. Human Reliability Analysis, 3.12.1. Outer Loadings Criteria 3.12.2. Reliability Criteria, 3.12.3. Convergent Validity Criteria 3.13. Moderation Testing 3.14. Ethical Considerations 3.15. Chapter Summary 4. DATA ANALYSIS AND RESULTS 4.1. Introduction 4.2. Demographic Profile of Respondents, 4.2.1. Age Distribution of Respondents 4.2.2. Gender Distribution of Respondents. 4.2.3. Education Level of Respondents, 4.2.4. Professional Background of Respondents 4.2.5. Smartphone Usage Duration 4.2.6. Previous Button Phone Usage 4.3. Measurement Model Assessment 4.3.1. Outer Loadings 4.3.2. Constructing Reliability and Validity 4.3.3. Discriminant Validity 4.4. Structural Model Assessment 4.4.1. Path Coefficients 4.4.2. Bootstrapping Results 4.4.3. R-Square Analysis 4.5. Moderation Analysis of Age 4.6. Hypothesis Summary 4.7. Discussion of Findings 4.7.1. Discussion of Al Features and Al Influence 4.7.2. Discussion of Psychological Factors and Al Influence 4.7.3. Discussion of Al Influence and Smartphone Adoption 4.7.4. Discussion of Age Moderation 4.8. Overall Interpretation of Results 4.9. Summary 5. Conclusion and Recommendations 5.1. Introduction 5.2. Summary of Major Findings, 5.2.1. Findings Related to Al Features, 5.2.2. Findings Related to Psychological Factors 5.2.3. Findings Related to AI Influence 5.2.4. Findings Related to Age Moderation 5.3. Conclusion Based on Research Objectives 5.4. Overall Conclusion 5.5. Theoretical Implications 5.6. Practical Implications 5.7. Recommendations for Smartphone Companies 5.8. Recommendations for Marketers 5.9. Recommendations for AI Developers 5.10. Limitations of the Study 5.11. Recommendations for Future Research 5.12. Conclusion References Appendix - Questionnaire Draft Section A: Demographics Section B: AI Features Section C: Psychological & Social Factors Section D: AI Influence Section E: Smartphone Adoption & Usage Appendix A: Questionnaire Coding Appendix B: Likert Scale Coding Appendix C: Data Cleaning Procedure. Abstract The dynamism of mobile phones that have emerged in the last ten years has greatly changed consumer behavior and patterns of communication. This paper examines the paradigm of changing the old type of button phones with limited durability and simple calling/texting capabilities to the new smartphones with sophisticated features and the Artificial Intelligence (AI) capabilities. The study examines how the use of Al capabilities like smart voice assistants, typing suggestions, facial recognition, personalized recommendations, and smart cameras have made a difference in consumer behavior and prompted smartphone usage. Social and psychological causes of this change, such as the change in lifestyle, the need to stay connected, and the tendency to be more dependent on digital services are also studied. It is suggested to use a mixed method approach, which will be based on quantitative data about the sales trends and consumer behavior and the qualitative data about user experiences and decision making. The results of this study will seek to demonstrate the role of Al as a significant component of influencing the uptake of mobile technologies as a factor that leads to the death of the button phone and the solidification of smartphones as part of our lives. Keywords: Artificial Intelligence, Consumer Behavior, Smartphone Adoption, Feature Phones, Technology Shift, Al influence, PLS-SEM.

Date Published:2026

Computer vision based Detection of household leftovers for recognition to reduce food waste in Pakistan

Sana, Sara [22L-6519]

Submitted in fulfillment of the requirements for the degree of Bachelor of science in Business analytics to the Department of Management sciences. Abstract List of Tables List of Figures List of Acronyms 1. Introduction 1.1 Background and Context 1.2 Problem Statement 1.2.1 Objective 1.2.2 Relevance and Impact 1.3 Identification of Gaps and Opportunities 1.3.1 Research Gaps 1.3.2 Opportunities for Further Research 1.4 Analytical Approach and Methodology (Overview) 1.5 Scope and Limitations 1.5.1 Scope 1.5.2 Limitations: 1.6 Expected Outcomes and Benefits 1.6.1 Technical Outcomes: 1.6.2 Practical Benefits: 1.6.3 Broader Impact: 1.7 Summary of Chapter 2. Review of Literature 2.1 Overview of Existing Research 2.2 Summary of Key Studies 2.2.1 Food Image Recognition Datasets 2.2.2 Multi-Label Ingredient Recognition and Segmentation 2.2.3 Recipe Retrieval using Image-Text Embeddings 2.3 Deep Learning and Convolutional Neural Networks in Food 2.4 Challenges and Biases in Food Image Classification 2.4.1 Domain Shift and Lighting Variability 2.5 Implications for the Current Study 2.6 Summary of Chapter: 3. Methodology 3.1 Research Framework 3.2 Data Source and Collection 3.2.1 Dataset Citation and Description 3.3 Data Privacy and Ethical Guidelines 3.3.1 GDPR and Local Privacy Compliance 3.3.2 Consent and Bias Mitigation Procedures 3.4 Data Preparation and Pre-processing 3.4.1 Data Cleaning and Handling of Missing Values 3.4.2 Image Resizing, Normalization, and Augmentation 3.4.3 Colour and Illumination Balancing 3.4.4 Set Up Your Environment and Install Dependencies: 3.5 Exploratory Data Analysis (EDA) 3.5.1 Dataset split verification: 3.6 Model Development and Training 3.6.1 Automated backup checkpoints: 3.6.2 Implementation 3.7 Model Evaluation and Comparison 3.7.1 Accuracy of Validation and training 3.7.2 Visualize Metrics-Radar Chart: 3.7.3 Inference on test images: 4. Testing Phase 4.1 Testing on real extracted image 4.2 Recipie Suggestion when user uploads a picture same as above: 4.2.1 Recipe 1: Quick Garlic Tomato Chicken Skillet 4.2.2 Recipe 2: Simple Cucumber Garlic Salad 5. Packages and Tools Used 5.1 Python Libraries and Frameworks 5.2 Data Manipulation Tools 5.3 Visualization Tools 5.4 Web Framework 5.5 Summary References. Pakistan is facing a devastating situation with respect to food wastage as nearly 31-36 MT of end-of-food-chain food waste is being thrown away each year, despite a high level of food insecurity and malnutrition across the country, there is a need for practical tools to increase sustainability in their day-to-day cooking. This report is a Computer Vision Based Detection of Household Leftovers for Recognition to Reduce Food Waste in Pakistan, which is an Al based system for detecting the food ingredients with the help of computer vision and machine learning. The main goal is to decrease food waste, allowing users to recognize and utilize that food that is available to create new and viable meals. We compiled a custom dataset comprising of 17672 images across 20 ingredient classes and divided it into training (12370), validation (3534), and test (1768) subsets. YOLOv8s was trained for 50 epochs on a Tesla T4 GPU with an image size of 416×416 pixels. The trained model achieved a mean Average Precision at IoU 0.50 (mAP@50) of 97.3%, precision of 95.0%, recall of 94.3%, and an F1-score of 94.7%. Detected ingredients are converted to culturally relevant recipe suggestions using a Gemini-powered generative Al back end. The outcomes are consistent with the efficacy of the system for its real-world use in minimizing household food waste. Keywords: Object Detection, YOLOv8, Food Waste Reduction, Computer Vision, Pakistani Cuisine, Sustainability, Recipe Recommendation, Household Leftovers.

Date Published:2026

A hybrid weighted sustainability scorecard for the ravi revierfort urban development project : Benchmaking saphire bay and chahar bagh against leen-nd v4.1, four UN SDGs, and pakistan's local priority layer

Tahir Mauz Ul Haq[22L-[6464]

Submitted in fulfillment of the requirements for the degree of Bachelor of science in Business analytics to the Department of Management sciences. Contents 1. Introduction. 1.1 Project Overview 1.2 Project Scope 1.3 Project Objectives. 1.4 Identified Options 1.5 Structure of the Report. 2. Literature Review. 2.1 Advisory Outcomes 2.2 Theoretical Foundations. 2.3 Review of Relevant Literature 2.3.1 Critique of USRs.. 2.3.2 Multi-Criteria Decision Analysis in Sustainability Assessment. 2.3.3 Urban Resilience and Climate Vulnerability 2.3.4 The Pakistani Sustainability Context.... 2.3.5 The Implementation Gap 2.3.6 The Cost-Benefit Frame for Sustainability Investment 2.4 Key Concepts and Definitions 2.5 Key Assumptions. 2.6 Constraints 3. Research and Analysis 3.1 Course of Action for Data Collection 3.2 Data Sources 3.3 Data Collection Methods 3.4 Sampling Approach 3.5 Data Storage and Management 3.6 Hybrid Framework Construction. 3.7 Analytic Hierarchy Process Application.. 3.8 Sensitivity Analysis Method 3.9 Conservative-Default Rule. 4. Review of findings and analysis of options 4.1 Data Validation 4.2 Headline Results 4.3 Sapphire Bay Detailed Scoring. 4.4 Chahar Bagh Detailed Scoring. 4.5 Comparative Analysis.... 4.6 Gap Analysis.... 4.7 Sensitivity Validation Results. 4.8 LEED-ND Direct Validation Exercise... 5. Selected Options and Rationale 5.1 Selection Criteria 5.2 Selected Interventions. 5.2.1 Quick Win Interventions. 5.2.2 High-Priority Interventions. 5.2.3 Strategic Interventions 5.2.4 Medium-Priority Interventions 5.3 Perceived Consequences. 5.3.1 Positive Outcomes 5.3.2 Negative Consequences and Mitigation. 5.4 Resource Assessment. 5.5 Budget Considerations. 6. Proposed Plan of Action 6.1 Action Steps and Implementation Roadmap.. 6.2 Milestones and Key Performance Indicators 6.3 Monitoring and Evaluation 6.4 Change Management 6.5 Contingency Planning. 7. Conclusions and Limitations.. 7.1 Project Recap 7.2 Key Findings. 7.3 Selected Recommendations Overview 7.4 Anticipated Benefits..... 7.5 Statement of Limitations. 7.5.1 Data Limitations... 7.5.2 Scope Limitations 7.5.3 Methodological Limitations. 7.5.4 External Factors 7.5.5 Resource Constraints 7.6 Recommendations for Future Research. References.... Appendix A: AHP Matrices and Eigenvector Calculations.. A.1 Sapphire Bay (Commercial-Mixed Typology) - Pairwise Comparison Matrix.... A.1.1 Eigenvector and Consistency Calculations. A.2 Chahar Bagh (Residential-Heritage Typology) A.2.1 Eigenvector and Consistency Calculations. A.3 Full Sensitivity Analysis Output. Appendix B: Public-Domain Source Inventory B.1 RUDA Institutional Documentation.. B.2 Pakistani Policy and Regulatory Documentation. B.3 International Benchmarks B.4 Academic Literature Sources. Appendix C: KPI Calibration Rubrics. C.1.1 KPI A1: Water Use Reduction. C.1.2 ΚΡΙ Α2: Non-Potable Water Reuse. C.1.3 KPI A3: Renewable energy integration C.1.4 KPI A4: Minimum Energy Performance C.2 Pillar B-Climate Resilience C.2.1 KPI B1: Climate Risk Mitigation (Flood). C.2.2 KPI B2: Green Space Index C.2.3 ΚΡΙ Β3: Urban Heat Island Reduction Pairwise Comparison Matrix.. C.3 Pillar C Urban Livability and Social Sustainability C.3.1 KPI C1: Compact Development Density C.3.2 KPI C2: Access to Quality Transit. C.3.3 KPI C3: Mix of Uses and Social Equity Appendix D: Stakeholder Feedback and Official Correspondence. D.1 Official Endorsement from RUDA.. D.2 Summary of Verbal Consultations. Abstract The largest planned riverfront megaproject in Pakistan is the Ravi Riverfront Urban Development Project (RRUDP). It seeks to lessen Lahore's burden from urbanisation. Measuring its sustainability is difficult, though, North American cities are the target audience for international rating systems such as LEED-ND v4.1. They frequently fall short of capturing the unique priorities of growing markets in South Asia. We developed a hybrid sustainability scorecard to close this gap. We used this technology on two RRUDP precincts: the 168.22-acre Chahar Bagh Phase II and the 5,000-acre Sapphire Bay. Four UN Sustainable Development Goals, a new Local Priority Layer, and LEED-ND v4.1 credits are all included in our framework. Pakistani laws, such as the RUDA Act 2020 and the Punjab Clean Air Action Plan, are the foundation of this local layer. To give our criteria particular weights, we employed the Analytic Hierarchy Process (AHP). We had excellent mathematical consistency. Additionally, we used a sensitivity analysis to validate the framework, and our rankings were 100% consistent across 15 test scenarios. We used a stringent "Conservative-Default Rule" to make sure we didn't overstate any ratings because some project data was confidential. Sapphire Bay received 50.8% (Silver Equivalent) and Chahar Bagh received 59.8% (Gold Equivalent) in the final results. We also used unaltered LEED-ND rules to score the precincts in order to demonstrate the value of our approach. Both projects failed to certify at all under the stringent international regulations. This demonstrates that our local layer effectively corrects the inherent bias in universal rating systems. Lastly, we developed a 10-step action plan that was prioritised. It identifies two regulatory "quick wins" that might enable both precincts to meet the Gold Equivalent criteria in less than 18 months. International green funding will be drawn in as a result. A scorecard.

Date Published:2026