New Titles

Asymmetric Cost Behavior and Financial Distress

Ahmed,Agha Mureed

Submitted to the Fast School of Computing in partial fulfillment of the requirement for the degree of Doctor of Philosophy in (Management Sciences) at the National University of Computer and Emerging Sciences. Chapter 1 1 Introduction 1.1 Background 1.1.1 Cost Behavior 1.1.2 Symmetric and Asymmetric Cost Behavior 1.1.3 Financial Distress 1.1.4 Asymmetric Cost Behavior, Financial Performance, and Financial Distress 1.2 Theoretical Justification 1.2.1 Conceptual Framework 1.3 Research Objectives and Questions 1.4 Scope of the Research 1.5 Significance 1.6 Motivation 1.7 Structure of the Dissertation Chapter 2 2 Literature Review 2.1 Cost Behavior 2.2 Theories of Cost Behavior 2.2.1 Fixed and Variable Cost 2.2.2 Mixed Cost Theory 2.2.3 Theory of Asymmetric Cost Behavior 2.3 Cost Categories 2.4 Empirical Models 2.5 Theories 2.5.1 Theory of Cost Asymmetry 2.5.2 Agency Theory 2.5.3 Resource Adjustment Cost Theory 2.5.4 Behavioral Finance Theory 2.6 Managerial and Operational Factors 2.6.1 Resource Adjustment 2.6.2 Managerial Expectations 2.6.3 Managerial Incentives, Corporate Governance, and Agency Problem 2.7 Cost Behavior and Decision-Making 2.7.1 Retaining Idle Resources and Cost Stickiness 2.7.2 Managerial Overconfidence and Optimism 2.7.3 Behavioral Biases in Downsizing Decisions 2.8 Financial Performance and Financial Distress 2.8.1 Financial Distress Measures 2.8.1.1 Altman (1968) 2.8.1.2 Ohlson (1980) 2.8.1.3 Zmijewski (1984) 2.8.1.4 Interpretation 2.8.1.5 Hillegeist et al. (2004) 2.8.1.6 Campbell et al. (2008) 2.9 Hypothesis Development 2.10 Strategies to Mitigate ACB's Impact 2.11 Chapter Summary Chapter 3 3 Research Methodology and Data 3.1 Introduction 3.2 Research Philosophies and Paradigms 3.2.1 Positivist Paradigm 3.3 Research Strategy 3.4 Sample Selection and Data Sources 3.5 Dependent Variable 3.5.1 Financial Performance 3.5.1.1 Return on Assets (ROA) 3.5.1.2 Earnings Power 3.5.1.3 Stock Returns 3.5.2 Financial Distress 3.5.2.1 ZM-Score 3.5.2.2 Z-Score 3.6 Independent Variable 3.6.1 Asymmetric Cost Behavior 3.7 Control Variables 3.7.1 Financial Leverage (Lev) 3.7.2 Newness (New) 3.7.3 Free Cash Holdings (Slack) 3.7.4 Asset Tangibility (Tang) 3.7.5 Depreciation (Dep) 3.7.6 Tobin's Q 3.7.7 Size 3.7.8 Dividend Yield 3.8 Empirical Model 3.8.1 Estimation Techniques and Justification 3.8.2 Ordinary Least Squares (OLS) Estimation 3.8.3 Limitations of Static OLS 3.8.4 Fixed Effects (FE) Estimation 3.8.5 Rationale for Choosing Fixed Effects (FE) over Random Effects (RE) 3.8.6 Diagnostic and Validity Tests 3.8.6.1 Multicollinearity 3.8.6.2 Heteroscedasticity 3.8.6.3 Autocorrelation Test 3.8.7 Reliability of Static Models (OLS and FE) 3.8.8 The Problem of Endogeneity 3.8.8.1 Reverse Causality 3.8.8.2 Dynamic Panel Data Estimation 3.8.8.3 Unobserved Heterogeneity 3.8.9 System GMM Approach for Dynamic Panel Data Analysis 3.8.10 Assessing the Reliability of System GMM in Dynamic Panel Models 3.8.10.1 Autocorrelation Testing: First-Order (AR1) and Second-Order (AR2) 3.8.10.2 Hansen J Test: Diagnosing Over-Identified Instruments 3.8.10.3 The Difference-in-Hansen Test of Exogeneity 3.8.10.4 Assumption of Steady-State 3.8.10.5 Instrument Count Method 3.9 Addressing Endogeneity in ACB and Financial Performance and Financial Distress Chapter 4 4 Empirical Results 4.1 Introduction 4.2 Descriptive Statistics 4.3 Pearson Pairwise Correlation 4.4 Study 1: Asymmetric Cost Behavior and Financial Performance 4.5 Empirical Model 4.6 Empirical Results 4.7 Robustness Test for ROA 4.7.1 Sample Attrition and Selection-on-Observables Analysis 4.8 ACB and Financial Performance During Global Financial Crisis 4.9 Addressing Endogeneity in ACB and Financial Performance 4.10 Discussion 4.11 Study 2: Asymmetric Cost Behavior and Financial Distress 4.12 Empirical Model 4.13 Empirical Results 4.14 Robustness Analysis 4.15 The Impact of Asymmetric Cost Behavior on Financial Distress 4.16 Impact During the Global Financial Crisis 4.17 Controlling for Endogeneity in the Effect of Cost Stickiness on Financial Distress 4.18 Conclusion Chapter 5 5 Conclusion 5.1 Purpose and Scope of the Study 5.2 Summary of Key Findings 5.3 Theoretical Contributions 5.4 Practical and Managerial Implications 5.5 Limitations and Future Research Directions. This dissertation examines the impact of asymmetric cost behavior (ACB) on firm financial performance and its role in increasing the likelihood of financial distress. Unlike traditional cost theories, which assume costs change proportionally with activity levels, ACB occurs when costs rise more sharply during growth than they decline during sales reductions. This behavior may be explained by managerial optimism, overconfidence, and agency-related incentives, which can contribute to rigid cost structures that reduce financial flexibility and increase vulnerability to financial distress. Using a final analytical dataset of 117,600 firm-year observations from 14,816 U.S. non-financial firms, with regression estimation samples varying according to complete-case availability spanning 1987-2020, the study investigates how cost stickiness, particularly in cost of goods sold (COGS) and selling, general, and administrative (SG&A) expenses, affects financial performance and distress. The analysis integrates behavioral finance, agency theory, and cost asymmetry theory to explain the mechanisms driving ACB, highlighting managerial decisions and cognitive biases as key contributors to cost rigidity. Empirical results show that firms exhibiting higher ACB retain excess resources during downturns, resulting in inflexible costs, reduced profitability, and increased risk of financial distress. The relationship between ACB and distress remains robust across various financial indicators (Altman Z-score, Ohlson O-score, Zmijewski score), and economic cycles. In addition, the findings highlight the potential importance of effective governance and managerial oversight in limiting behaviors that contribute to persistent cost stickiness and financial vulnerability. The dissertation contributes theoretically by linking ACB to financial distress, expanding the understanding of cost behavior beyond accounting effects to its strategic and behavioral implications. Practically, the findings provide guidance for managers, investors, creditors, and policymakers on mitigating the financial risks associated with cost stickiness, emphasizing the need for proactive cost management, behavioral awareness, and enhanced governance mechanisms. Keywords: Asymmetric Cost Behavior, Cost Stickiness, Financial Performance, Financial Distress, Managerial Bias, Behavioral Finance, Corporate Governance.

Date Published:2026

Mitigate or Intensify : Board Cooption and Bank Risk Culture.

Khalil Ur Rahman

Submitted to the Fast School of Computing in partial fulfillment of the requirement for the degree of Doctor of Philosophy in (Management Sciences) at the National University of Computer and Emerging Sciences. 1 Introduction 1.1 Board Cooption and Bank Risk Culture 1.1.1 Board Cooption and Liquidity Risk 1.1.2 Board Cooption and Credit Risk 1.1.3 Board Cooption and Solvency Risk 1.2 Significance of the Study 1.3 Study 1: Board Dynamics and Risk Management: Cooption and Its Influence on Liquidity Risk in Australia and New Zealand Banks 1.4 Study 2: Deciphering Hidden Nexus: Board Cooption's Impact on Credit Risk in the Australian and New Zealand Banks 1.5 Study 3: Board Governance and Solvency Risk: The Role of Coopted Directors 1.6 Rationale of the Study 1.7 Gap Identification 1.8 Develop a Holistic Understanding of the Study Phenomena 1.9 Integration of Different Theoretical Perspectives 1.10 Use of Different Analytical Techniques 1.11 Scope of the Dissertation 1.12 Alignment of Dissertation with United Nations Sustainable Development Goals (SDGs) 1.13 Structure of the Dissertation 1.14 Chapter Summary Chapter 2 2 Literature Review 2.1 Introduction 2.2 Board Dynamics and Risk Management: Cooption and Its Influence on Liquidity Risk in Australia and New Zealand Banks 2.2.1 Board Cooption: Effective Measure of Corporate Board Governance 2.2.2 Bank Risk Management 2.2.3 Relationship of Board Cooption and Liquidity Risk 2.2.4 Hypothesis Development 2.3 Deciphering Hidden Nexus: Board Cooption's Impact on Credit Risk in the Australian and New Zealand Banks 2.4 Board Cooption: An Effective Measure of Corporate Governance (Credit Risk) 2.4.1 Significance of Corporate Governance in Bank Risk Management 2.4.2 Relationship of Board Cooption and Credit Risk 2.4.3 Hypothesis Development 2.5 Board Governance and Solvency Risk: The Role of Coopted Directors 2.5.1 Board Cooption: An Effective Measure of Board Governance 2.5.2 Bank Risk Management 2.5.3 Relationship of Board Cooption and Solvency Risk 2.5.4 Hypothesis Development 2.6 Banking Fraud Instances Reported in Australia and New Zealand 2.7 Consolidated Theoretical Framework of the Dissertation Chapter 3 3 Data and Methodology 3.1 Introduction 3.2 Research Philosophy 3.3 Research Strategy 3.4 Study 1: Board Dynamics and Risk Management: Cooption and Its Influence on Liquidity Risk in Australia and New Zealand Banks 3.4.1 Population and Sample Statistics 3.5 Study 2: Deciphering Hidden Nexus: Board Cooption's Impact on Credit Risk in the Australian and New Zealand Banks 3.5.1 Population and Sample Statistics 3.6 Study 3: Board Governance and Solvency Risk: The Role of Coopted Directors 3.6.1 Population and Sample Statistics Chapter 4 4 Analysis and Discussion 4.1 Introduction 4.2 Study 1: Board Dynamics and Risk Management: Cooption and Its Influence on Liquidity Risk in Australia and New Zealand Banks 4.2.1 Robustness Check 4.2.2 Recommendations and Practical Implications 4.3 Study 2: Deciphering Hidden Nexus: Board Cooption's Impact on Credit Risk in the Australian and New Zealand Banks 4.3.1 Robustness Check 4.3.2 Recommendations and Practical Implications 4.4 Study 3: Board Governance and Solvency Risk: The Role of Coopted Directors 4.4.1 Robustness Check 4.4.2 Recommendations and Practical Implications Chapter 5 5 Conclusion 5.1 Introduction 5.2 Study 1: Board Dynamics and Risk Management: Cooption and Its Influence on Liquidity Risk in Australia and New Zealand Banks 5.2.1 Limitations and Future Directions 5.3 Study 2: Deciphering Hidden Nexus: Board Cooption's Impact on Credit Risk in the Australian and New Zealand Banks 5.3.1 Limitations and Future Directions 5.4 Study 3: Board Governance and Solvency Risk: The Role of Coopted Directors 5.4.1 Limitations and Future Directions 5.5 Consolidated Conclusion 6 References. This dissertation articulates three purposefully designed studies regarding Board Cooption and Bank Risk Culture. A detailed analysis of how an increased number of coopted directors on bank governance boards affects bank risk management drawing upon risk transformation, Agency theory, and financial intermediation theories. The bank risk management has been assessed through three significant risks, as are also highlighted BASEL Banking Accords, le. Liquidity risk, Credit Risk, and Solvency Risk, which also form the segregation bases of three studies. The first study investigates the relationships of Board Cooption and Liquidity Risk in Australian and New Zealand banks. The sample data consists of banks from both countries and data for the period from 2011 to 2021. The empirical results demonstrate that a higher proportion of coopted directors on the board are significantly associated with increased liquidity risk. These findings exhibit robustness across the measures of Board Cooption, Tenure Weighted Cooption, and Residual Cooption, thereby substantiating the applicability of the econometric model. Furthermore, a distinctive comparative analysis based on separate subsamples for both countries yields results that are consistent with the full sample, thereby reinforcing the econometric model. The second study examines how board.cooption impacts credit risk within banks of Australia and New Zealand. Using sample data similar to first study, the results obtained through fixed effect and random effect regression models reveal that an increased proportion of coopted directors on the board is associated with a notably increased value of the Loan Loss Reserve ratio, signifying an elevated level of credit risk for banks. The evidence is consistent with the notion that coopted directors bring about less effective board monitoring, which escalates agency problems, thereby enhancing credit risk. The results are robust to alternative measures of credit risk, called the Loan Loss Provision ratio, and also with GMM regression for all cooption measures. The third study investigates the relationship between solvency risk and board cooption for the same sample banks' data. The study finds that an increased proportion of coopted directors on the board is associated with a notable increase in solvency risk for banks, which also develops agency problems and ineffective management of solvency risk. This dissertation contributes significantly to the literature by highlighting the fact that banks in Australia and New Zealand should develop robust risk management frameworks under the Basel Accords for managing major risks through effective Board Governance. This dissertation also significantly contributes by emphasizing the need to actively monitor the appointment and performance of board members in Australian and New Zealand banks. Additionally, this dissertation suggests that regulators should carefully monitor board cooption as a sensitive measure of board governance to ensure effective risk management policies. Moreover, the regulators in the financial services industry may devise mechanisms to regulate banks and avoid risk management issues. The study is considered a pioneering effort and provides fresh insight as a reference point for future researchers. Keywords: Board Cooption, Bank, Liquidity Risk, Credit Risk, Solvency Risk, Risk Management, Governance, BASEL Accords, Australia, New Zealand.

Date Published:2026

Development of empirical relations for setting time and compressive strengths

Janjua, Nabeel Ahmad [22L-7161]

Submitted in fulfillment of the requirements for the degree of BS in Civil Engineering to the Department of Civil Engineering. Table of Contents CERTIFICATE OF APPROVAL ACKNOWLEDGEMENT ABSTRACT List of Figures List of Tables Chapter 1 Introduction 1.1 Brief Overview of Concrete and Its Elements 1.2 Cement Paste as the Governing Phase 1.3 Early-Age Behaviour and the Motivation for Prediction 1.4 The 28-Day Strength Convention and the Cost of Waiting 1.5 Problem Statement 1.6 Research Objectives 1.7 Scope and Limitations 1.8 Significance of the Study 1.9 The Global Context of Cement and Concrete 1.10 From Fresh Suspension to Load-Bearing Solid 1.11 The Rationale for Early-Age, Non-Destructive Prediction 1.12 Overview of the Adopted Early-Age Indicators 7 1.13 The Engineering Significance of Compressive Strength 1.14 Organisation of the Report Chapter 2 Literature Review 2.1 Historical Development: From Vicat to Multi-Sensor Prediction 2.1.1 Foundations (1818–1824) 2.1.2 Chemistry and Empiricism (1887–1929) 2.1.3 Maturity and Kinetics (1946–1977) 2.1.4 Microstructure and Non-Destructive Monitoring (1990s–2000s) 2.1.5 Multi-Sensor and Data-Driven Prediction (2010s–present) 11 2.2 Cement Hydration Chemistry 2.3 Abrams’ Law and the Water-to-Cement Ratio 2.4 The Interfacial Transition Zone 2.5 Setting Time and the Vicat Method 2.7 Electrical Resistivity and Microstructural Development 2.8 Autogenous Shrinkage in Low w/c Systems 2.9 Ultrasonic Pulse Velocity 2.10 Simultaneous Testing of Early-Age Properties 2.11 Empirical and Data-Driven Models for Setting Time and Strength 2.12 Research Gap 2.13 Portland Cement: Composition and Manufacture 2.14 The Chemistry of Cement Hydration 2.15 Kinetics and the Periods of Hydration 2.16 Microstructure Development and the Role of C-S-H 2.17 Porosity, Gel-Space Ratio and the Strength-Porosity Relationship 2.18 Maturity Methods 2.19 Electrical Resistivity of Cementitious Systems 2.20 Calorimetry of Cement Hydration 2.21 Empirical and Theoretical Strength-Age Models 2.22 Autogenous Shrinkage and Self-Desiccation 2.23 Influence of Temperature on Hydration Kinetics 2.24 Fineness and Particle Size Distribution 2.25 Ultrasonic Pulse Velocity and Elastic Wave Propagation 2.26 Supplementary Cementitious Materials and Blended Systems 2.27 Chemical Admixtures and the Control of Early-Age Behaviour 2.28 Statistical and Data-Driven Strength Models 2.29 Durability Implications of the Early-Age Microstructure 2.30 The Degree of Hydration and Its Determination 2.31 Pore Solution Chemistry and Its Evolution 2.32 Characterisation of the Pore Structure 2.33 Compressive Strength Testing: Standards and Specimen Effects 2.34 Chemical Shrinkage and the Water Demand of Hydration 2.36 Mathematical Models for the Progress of Hydration 2.37 Summary of the Literature and Positioning of the Present Work Chapter 3 Methodology 3.1 Experimental Philosophy 3.2 Materials and Mix Proportions 3.3 Specimen Preparation 3.4 Vicat Setting Time (ASTM C191) 3.5 Calorimetry (Heat of Hydration) 3.6 Autogenous Shrinkage 3.7 Electrical Resistivity 3.8 Ultrasonic Pulse Velocity 3.9 Compressive Strength Testing 3.10 Data Analysis and Model Development 3.11 Theoretical Basis of the Early-Age Indicators 3.12 The Hydration Time Constant from Calorimetry 3.13 The Zero Time from Electrical Resistivity 3.14 Formulation of the Empirical Strength Relation 3.15 Estimation of the Parameters by Least Squares 3.16 Validation Methodology 3.17 Formulation of the Comparison Models 3.18 Calibration and Standardisation of the Instruments 3.19 Repeatability, Uncertainty and Error Analysis 3.20 Sensitivity of the Calibrated Parameters 3.21 Assumptions and Their Justification 3.22 Worked Example of the Strength Prediction 3.23 Summary of the Analytical Procedure 3.24 Statistical Measures Used in the Assessment Chapter 4 Results and Discussion 4.1 Heat of Hydration 4.2 Setting Time 4.3 Electrical Resistivity and the Zero Time 4.4 Autogenous Shrinkage 4.5 Ultrasonic Pulse Velocity 4.6 Compressive Strength Development 4.7 Correlation of Early-Age Kinetics with Strength Development 4.8 Development of the Empirical Relation 4.9 Validation against the 28-Day Strength 4.10 Comparison with Established Models 4.11 Limitations Chapter 5 Conclusions and Recommendations 5.1 Conclusions 5.2 Practical Recommendations for Quality Control 5.3 Recommendations for Future Work References List of Figures Fig. 1.1 Typical heat-evolution rate of a hydrating Portland cement paste, showing the five characteristic periods of hydratio Fig. 2.1 Historical development of cement setting and strength prediction, from the foundational work of the early nineteenth Fig. 2.2 Illustrative Abrams-type relationship between the water-to-cement ratio and the resulting compressive strength, marki Fig. 2.3 Illustrative relationship between capillary porosity and compressive strength of hardened cement paste, showing the s Fig. 2.4 The maturity concept: compressive strength expressed as a single function of the equivalent age at a reference temper Fig. 2.5 Characteristic evolution of electrical resistivity with time for a hydrating cement paste, showing the early decrease Fig. 2.6 Comparison of the principal functional forms used to model strength development with age: the unbounded logarithmic f Fig. 2.7 Schematic influence of curing temperature on strength development, showing faster early strength gain but a tendency Fig. 2.8 Schematic particle-size distributions for cements of different fineness; finer cements expose more surface area and h Fig. 2.9 Illustrative development of ultrasonic pulse velocity with time, rising as the paste stiffens and saturating once a c Fig. 2.10 Heat-based degree of hydration of the studied paste, obtained by normalising the cumulative thermal signal by its ult Fig. 3.1 Steel prism moulds used to cast companion specimens for compressive strength testing. Fig. 3.2 Vicat apparatus with the needle penetrating the cement paste for setting-time determination (ASTM C191). Fig. 3.3 Calorimeter cells (Calmetrix F-Cal) used to record the heat of hydration of the paste. Fig. 3.4 Length-comparator shrinkage rig used to record early-age autogenous length change of the sealed paste. Fig. 3.5 Electrical resistivity test arrangement: DC power supply, paste cell, and multimeter. Fig. 3.6 Electrical resistivity measurement on the moulded paste specimen during early hydration. Fig. 3.7 Ultrasonic transducers clamped to the paste specimen for pulse-velocity measurement. Fig. 3.8 Ultrasonic pulse velocity instrument used during early-age monitoring Fig. 3.11 Linearised form of the empirical relation: the measured calibration strengths plotted against the transformed age X f Fig. 3.12 Residuals of the calibrated relation at each calibration age Fig. 3.13 Comparison of the empirical relation developed in this work with a Plowman fit to the same calibration data Fig. 3.14 Sensitivity of the fitted slope b to the choice of calibration data Fig. 4.1 Calorimetry heat-rate signal (sample–reference temperature diff Fig. 4.2 Cumulative heat-equivalent maturity versus time with the fitted Freiesleben Hansen–Pedersen hydration-kinetics curve. Fig. 4.3 Vicat needle penetration versus time, indicating initial set at approximately 200 min and final set at approximately Fig. 4.4 Electrical resistivity versus time, with the minimum (zero time) at approximately 40 min marking percolation onset. Fig. 4.5 Autogenous length change of the sealed paste versus time, reaching approximately −449 µm at 48 hours. Fig. 4.6 Compressive strength versus age (log scale) with the fitted empirical model and the held-out 28-day validation point. Fig. 4.7 Calibration of the strength models on data up to 15 days and validation against the held-out 28-day strength (measure. . The prediction of compressive strength and setting time at early ages is a critical aspect of concrete technology, as it directly influences construction scheduling, quality control, and structural safety. Conventional strength assessment relies on destructive testing at 28 days, which provides delayed feedback. This research develops an empirical relationship that estimates the 28-day compressive strength of cement paste from early-age, largely non-destructive measurements completed within the first two days. Ordinary Portland Cement paste at a water-to-cement (w/c) ratio of 0.27 was prepared under controlled laboratory conditions (25–27 °C). A suite of tests was conducted simultaneously on specimens from the same batch so that elapsed time could serve as a common reference axis: setting time by the Vicat apparatus (ASTM C191); heat of hydration by calorimetry; autogenous length change (shrinkage); electrical resistivity, from which the “zero time” of percolation was extracted; and ultrasonic pulse velocity. Compressive strength was measured destructively at ages from 24 hours to 28 days. The calorimetry curve was fitted to a hydration-kinetics model, and the resistivity minimum and Vicat setting time were used to define the physically meaningful time origin and time scale of strength development. An empirical relation of logarithmic form was calibrated on strengths up to 15 days, deliberately excluding the 28-day result. When validated against the held-out 28-day strength of 69.9 MPa, the model predicted 74.3 MPa, an error of approximately +6 % and within the bootstrap 95 % confidence interval. A key finding is that heat-equivalent hydration saturates within about two days while strength continues to rise, so electrical resistivity and calorimetry are more informative for early hydration monitoring than ultrasonic pulse velocity for this paste. The results confirm that early-age testing can estimate 28-day compressive strength without waiting 28 days, providing a practical tool for quality control. The principal limitation is that calibration was performed for a single w/c ratio; extension to multiple ratios is recommended as future work. Keywords: Cement paste; Compressive strength; Setting time; Calorimetry; Electrical resistivity; Empirical relation; Early-age prediction. .

Date Published:2026

Transition from conventional to autonomous vehicles : a study at an urban signalized intersection

Zabeeh, Areej [22L-7193]

Submitted in fulfillment of the requirements for the degree of BS in Civil Engineering to the Department of Civil Engineering. CERTIFICATE OF APPROVAL ACKNOWLEDGEMENT ABSTRACT LIST OF FIGURES LIST OF TABLES LIST OF ABBREVIATIONS CHAPTER 1 INTRODUCTION 1.1 Background of Study 1.2 Driving the future: Global Emergence of Autonomous Vehicles 1.3 Study Area 1.4 Problem Statement 1.5 Study Objectives 1.6 Scope of the Study 1.7 Significance of Study 1.8 Relevance of the Study to Sustainable Development Goals (SDGs) 1.9 Thesis Overview CHAPTER 2 LITERATURE REVIEW 2.1 Evolution and Technological Development of Autonomous Vehicles 2.2 Feasibility of Autonomous Vehicle Implementation in Developing Countries 2.3 Theoretical and Strategic Perspectives on Technological Transition 2.4 Managing the Transition from Conventional to Autonomous and Electrical Vehicles 2.5 Simulation-Based Evaluation of Autonomous Vehicles at Signalized Intersections 2.6 Impact of Autonomous Vehicles on Traffic Flow and Capacity 2.7 Mixed Traffic Conditions and Challenges in Urban Environments 2.8 Autonomous Driving Research Platforms and System Architecture 2.9 Autonomous Vehicle Feasibility Studies and Infrastructure Readiness 2.10 Use of PTV Vissim in Traffic Simulation 2.11 Earlier Calibration of PTV Vissim for Traffic Simulation 2.12 Simulation of Cooperative Automated Vehicles (CAVs) 2.13 Simulation of Autonomous Automated Vehicles (AAVs CHAPTER 3 METHODOLOGY 3.1 Study Site and Survey Parameters 3.2 Data Collection 3.2.1 Traffic Volume and Vehicle Composition 3.2.2 Signal Timings 3.2.3 Vehicle Speeds 3.3 Model Development in PTV Vissim 3.3.1 Model Development Procedure 3.4 Calibration of the Model 3.4.1 Driving Behavior 3.4.2 Calibration Trials 3.5 Assignment of Different Driving Logic of Autonomous Vehicles 3.5.1 Establishing Node 3.5.2 Introducing Different Driving Logics 3.5.3 Simulation Scenarios 3.5.4 Simulation Runs and Extraction of Results CHAPTER 4 RESULTS AND DISCUSSION 4.1 Results 4.1.1 Average Vehicle Delay 4.1.2 Stop Delay 4.1.3 Queue Length 4.1.4 Combined Summary 4.2 Discussion 4.2.1 Overview 4.2.2 Network Delay Falls While Intersection Delay Rises 4.2.3 Why Driving Behavior Matters So Much 4.2.4 Queue Length Dynamics 4.2.5 Low Penetration Behavior: The Transitional Range 4.2.6 Implications for Traffic Management in Lahore 4.2.7 Comparison with Prior Literature 4.2.8 Limitations and Future Research Directions 4.2.9 Summary CHAPTER 5 CONCLUSION AND RECOMMENDATIONS 5.1 Conclusion 5.2 Recommendations REFERENCES. Urban signalized intersections remain among the most congested and accident-prone points in any road network. The gradual introduction of autonomous vehicles (AVs) into conventional traffic presents both an opportunity and a challenge that demands careful evaluation before widespread adoption can be considered practical. The data of traffic volume, vehicle composition, speed, and signal timing have been collected in the field to develop a microscopic traffic simulation model in PTV VISSIM. Three types of AV driving behavior, including cautious, normal, and aggressive, were studied at 0%, 10%, 20%, 40%, 60%, 80%, and 100% of AV penetration rates. However, the results show that the increase in the proportion of autonomous vehicles leads to better performance at intersections; nevertheless, the level of benefit depends on the behavior type chosen. Aggressive behavior was the most beneficial in terms of performance among all tested behaviors, since it decreased the average vehicle delay from 88.49 s/vehicle in traditional traffic to 81.76 s/vehicle at 100% of the autonomous vehicles, a decrease of 7.6%. The benefits were especially evident above 40% of AV penetration, while low proportions of AVs (10-20%) resulted in minor changes in performance due to interaction with traditional vehicles. Cautious behavior resulted in the lowest performance improvement. The results obtained from the study are quite helpful in offering some valuable lessons to planners and policymakers on how to adopt the integration of autonomous vehicles into the mix of traffic flow in developing countries. This is because the results highlight the importance of selecting an effective AV driving strategy to enhance traffic efficiency. Keywords: Autonomous Vehicles (AVs), Conventional Vehicles, Signalized Intersection, Driving Behavior, Cautious Mode, Normal Mode, Aggressive Mode, Traffic Flow, Simulation.

Date Published:2026

Impact of petroleum prices on staple consumer products in Lahore

Ghazanfar, Maheen [22L-6511]

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 of the problem 1.2. Research Objectives.. 1.3. Scope of the study 2. Data Collection and Descriptive Analysis 2.1. Introduction 2.2. Data Collection Methodology 2.3. Products Selected 2.4. Petrol Price Overview 2.5. Descriptive Statistics. 2.6. Neighbourhood Comparison. 2.7. Monthly Trend Analysis (Month on Month Averages) 2.8. Key Findings. 3. Statistical Analysis 3.1. Introduction. 3.2. Regression Methodology 3.3. OLS Model. 3.4. Regression Results. 3.5. Interpretation of Each Product 3.6. Forecasting. 4. Policy Discussions and Recommendations. 4.1. Impact of Petroleum Prices and Sensitive Products 4.2. Government Policies 4.3. Priority Products 4.4. Practical Recommendations. 5. Conclusion 6. References. . This paper explores the transmission mechanism of macroeconomic price volatility of petroleum products into the price of essential consumer staples at retail level in Lahore. The study examines the retail markets prices of the five most important commodities namely Sunridge Atta (5kg), Rice (1kg), Sugar (1kg), Daal Mash Washed (500g), and Eggs (1 dozen) in two leading municipal markets of Wapda Town and Johar Town during the period of January to July 2026. The study uses descriptive statistical analysis and Ordinary Least Squares (OLS) regression modeling to quantify the price elasticity of these staples with respect to the price movements of Motor Spirit (Petrol) and High-Speed Diesel (HSD). The empirical findings show that there is a stark contrast in market reaction between logistics-intensive dry commodities, especially Sunridge Atta and Rice, which have a statistically significant, direct positive reaction to fuel and freight cost-push shocks. On the opposite side, perishables like Eggs shows a negative, non-significant correlation with the petroleum prices, and suggests a true decoupling due to the biological and seasonal cycle of supply. In addition, spatial retail market comparisons indicate that the Johar Town has artificial retail premiums for most items in the commodity basket. The study shows that the precarious condition of urban food security has exposed it to compounding logistics inflation, with the incorporation of baseline and geopolitical shock forecasting models till October 2026. Finally, the research finds that universal macroeconomic fuel support is sub-optimal and suggests targeted support for agricultural freight in the form of agricultural fuel subsidies for vulnerable dry staples, and localized support in the form of strict enforcement of price parity among competing fuel sources in the municipalities. Keywords: Staple Goods, Statistical Analysis, Descriptive Analysis, Regression, Price Fluctuations.

Date Published:2026

Zestora

Ahad, Muhammad Abdul [22L-7852]

Submitted in fulfillment of the requirements for the degree of BS in Software Engineering to the Department of Software Engineering. List of Figures List of Tables 1 Introduction 1.1 Purpose of this Document. 1.2 Intended Audience 1.3 Definitions, Acronyms, and Abbreviations 1.4 Conclusion 2 Project Vision 2.1 Problem Domain Overview 2.2 Problem Statement 2.3 Problem Elaboration 2.4 Goals and Objectives 2.5 Project Scope 2.6 Sustainable Development Goal (SDG) 2.7 Constraints 2.8 Business Opportunity 2.9 Stakeholders Description/ User Characteristics. 2.9.1 Stakeholders Summary 2.9.2 Key High-Level Goals and Problems of Stakeholders 2.10 Conclusion - Literature Review / Related Work 3.1 Definitions, Acronyms, and Abbreviations 3.2 Detailed Literature Review 3.2.1 Commercial Automation Platforms 3.2.2 Multi-Channel Messaging, Conversational Automation, and Data Integration 3.3 Related Work Summary Table 3.4 Conclusion 4 Software Requirement Specifications 4.1 List of Features 4.2 Functional Requirements 4.2.1 Authentication and User Management 4.2.2 Core Engine 4.2.3 WhatsApp Bot Automation 4.2.4 Telegram Bot Automation 4.2.5 Email Summarizer Automation 4.2.6 YouTube Comment Analyzer Automation 4.2.7 Google Maps Lead Generator Automation 4.3 Quality Attributes 4.4 Non-Functional Requirements 4.5 Assumptions 4.6 Use Cases 4.7 Hardware and Software Requirements 4.7.1 Hardware Requirements 4.7.2 Software Requirements 4.8 Graphical User Interface 4.9 Database Design 4.9.1 ER Diagram 4.9.2 Data Dictionary 4.10 Risk Analysis 4.10.1 Technical Risks 4.10.2 Security and Compliance Risks 4.10.3 Operational and Business Risks 4.11 Conclusion 5 High-Level and Low-Level Design 5.1 System Overview 5.2 Design Considerations 5.2.1 Assumptions and Dependencies 5.2.2 General Constraints. 5.2.3 Goals and Guidelines 5.2.4 Development Methods 5.3 System Architecture. 5.3.1 Subsystem Architecture 5.4 Architectural Strategies 5.4.1 Modular and Extensible Architecture 5.4.2 Core Automation Engine Design 5.4.3 API-Centric Integration 5.4.4 Data Management and Persistence 5.4.5 Error Handling and Recovery 5.4.6 User Interface Paradigm 5.4.7 Security and Authentication 5.4.8 Future Scalability and Maintainability 5.5 Domain Model/Class Diagram 5.6 Policies and Tactics 5.6.1 Coding Standards and Conventions 5.6.2 Use of Design Patterns 5.6.3 Testing and Quality Assurance 5.6.4 Requirements Traceability 5.6.5 Maintenance and Deployment 5.7 Conclusion 6 Implementation and Test Cases 6.1 Implementation 6.1.1 Frontend 6.1.2 Telegram Bot Automation 6.1.3 YouTube Comment Analyzer Automation 6.1.4 Email Automation 6.1.5 Google Maps Lead Generator 6.1.6 WhatsApp Bot Automation 6.2 Test Case Design and Description 6.3 Test Metrics 6.3.1 Functional Tests 6.4 Conclusion 7 User Manual 7.1 Authentication. 7.1.1 Sign-Up. 7.1.2 Login 7.2 Dashboard 7.2.1 Viewing the Dashboard Overview 7.3 Workflow Management 7.3.1 Viewing the Workflows List 7.3.2 Creating a New Workflow 7.3.3 Deleting or Managing Workflows 7.4 Workflow Editor (Automation Setup). 7.4.1 Navigating the Editor Interface. 7.4.2 Adding Nodes to the Canvas 7.4.3 Connecting Nodes 7.4.4 Configuring Nodes (Properties Panel) 7.4.5 Saving Workflows 7.4.6 Using the AI Agent Chat Panel. 7.5 Execution History 7.5.1 Viewing Executions Log 7.5.2 Debugging Executions 7.6 Settings and Configuration 7.6.1 Managing Credentials 7.6.2 Editing or Deleting Credentials 7.7 Conclusion 8 Conclusion and Future Work 8.1 Conclusion 8.2 Future Work. Zestora is a modular workflow automation platform designed to help small and medium-sized busi nesses and digital agencies automate their communication, lead generation, and customer engagement processes without writing code. The system provides an intuitive, no-code visual workflow builder that connects triggers, conditions, and actions, enabling users to automate tasks such as responding to messages, extracting leads, analyzing sentiment, and summarizing emails from multiple communication channels. This project is targeted at resolving the issue that a lot of companies have to deal with when they use various platforms for their communication, such as Whatsapp, Telegram, Youtub, Email, and Google Maps. The reason is that everything is managed manually. They typically overlook important leads, give late responses, or become disorganized. To overcome this, Zestora unites all these communication modes into one system through their APIs and then utilizes AI to deal with the unstructured messages. This allows companies to reply more quickly, improve their productivity, and be certain that no possible lead is ignored. This project introduces a visual workflow automation engine The users are not required to write any code. Rather, they can make workflows by darg-n-drop. The nodes can be used as triggers, conditions, and actions. The workflows decides how the incoming data (messages, comments, or listings) would be processed and whether the system would take actions such as replying, creating a new lead, or sum marizing the content. Besides WhatsApp Cloud API, Telegram API, YouTube Data API, Google Maps API, Gemini API, and email through IMAP, the platform is also compatible with various other chan nels. Additionally, it supports Slack and Google Workspace tools such as Sheets and Docs to facilitate reporting and team collaboration.. The platform includes several built-in automation modules: • Whatsapp and Telegram bots for automated responses, FAQs and lead extraction. • Email summarizer to summarize long conversations and forward concise updates. • YouTube comment analyzer to perform sentiment analysis and identify potential leads. • Google Maps Lead Generator to extract and structure business information. Every workflow execution is recorded, audit-proof, and can be exported in different formats (CSV, Excel, JSON), thus ensuring traceability. Zestora’s architecture is built around the quality attributes of scalability, security, and reliability. The system provides not only encrypted data storage but also extensibility for future integrations. .

Date Published:2026

Impact of transforming signalised corridor into signal free corridor on traffic operations and safety : a scenario-based study

Muhammad, Ali [24L-8423]

Submitted in fulfillment of the requirements for the degree of MS in Civil Engineering to the Department of Civil Engineering. . Certificate of Approval Declaration Plagiarism Certificate Abstract Dedication Acknowledgements Table of Contents List of figuresx List of tables Abbreviations Chapter 1 Introduction 1.1 Context and Background 1.2 The Lahore Context 1.3 Problem Statement 1.4 Research Objectives 1.5 Scope Definition 1.6 Thesis Organization Chapter 2 Literature Review 2.1 Introduction 2.2 Urban Traffic Management and Signalized Intersections 2.3 Grade Separation and Signal-Free Corridor Design 2.4 Signal-Free Corridors in Pakistan 2.5 Microsimulation as evaluation tool 2.6 Road Safety Assessment 2.7 Summary of Key Literature 2.8 Conclusion Chapter 3 Research Methodology 3.1 Introduction 3.2 Research Design and Overall Framework 3.3 Study Area Identification and Selection 3.4 Data Collection 3.5 Model Development in PTV Vissim 3.6 Simulation Run Parameters (Warmup Period, Simulation Duration and Run) 3.7 Model Calibration 3.8 Model Validation 3.9 Safety Assessment Framework 3.10 Measure of Effectiveness of Operational Analysis Chapter 4 Results and Discussions 4.1 Introduction 4.2 Traffic Operations 4.3 Road Safety Assessment 4.4 Integrated Discussion Chapter 5 Conclusion and Recommendations 5.1 Traffic Operations 5.2 Safety Assessment 5.3 Policy and Planning Recommendations 5.4 Study Limitations and Future Directions Chapter 6 References Appendix A – Rescue 1122 Data Appendix B - PTV Vissim (License Issuance) Appendix C - Traffic Count Summaries. The transformation of signalized corridors into signal free corridors (SFCs) through elimination of conventional control points and addition of grade separated structures and U-turns defines one of the most capital investment infrastructure developments in developing countries. Despite adoption of this popular trend in South Asian mega cities the net impact on traffic operational benefits and road safety remains empirical unresolved. The traffic conditions in these mega cities possess heterogenous traffic mix with weaker lane discipline as compared to developed countries. This study covers this gap and quantifies the net impact through scenario-based microsimulation comparison of signalized and signal-free Khayaban-e-Firdousi, Lahore. 4.2 km long section from Shouq Chowk to Shoukat Khanum Memorial Hospital (SKMH) Flyover was signalized corridor with three intersections and a roundabout in 2021 and all these control points were replaced with U-turns in 2024 transforming into signal-free corridor. Two microsimulation models were developed in PTV Vissim 2026 under equivalent 2024 forecasted peak hour traffic demand. Vissim follows Wiedemann-74 car following model for urban corridors, the model was calibrated longitudinally and laterally across all vehicle classes to reflect the heterogenous traffic mix and weak lane discipline of Lahore traffic streams. Model validation was carried out by conducting Journey Time survey using floating car method in PM peak hour and yielded average travel time of 421 seconds, the microsimulation model travel time results were within 15% acceptable deviation range. In total eleven simulation runs per scenario were conducted with total duration of 4,500 seconds with 900 seconds warmup period. The operational performance results revealed asymmetry on the corridor, the westbound direction shows minute reduction in travel time and significant reduction in average delay by 45.1%. The westbound direction depicts a contradictory output, a marginal increase in delays of 2.74 seconds and 5.3% increase in travel time. This asymmetry and deterioration of delay and travel time which is against the popular version of signal free corridor is due to the indued delay at MB Johar Town intersection replaced by a U-turn, it possesses high right turning volume as compared to other intersection. The average operating speed across all vehicle classes and links on the SFCs increases by 7.45 km/h, as compared to signalized corridors. Surrogated Safety Assessment by using FWHA SSAM, with TTC value of 0.75 seconds justified for similar traffic conditions revealed 11.2 % increase in predicted rear-end conflicts linked with higher operating speed and 8 % in total conflicts. The empirical safety analysis was carried out at macro level on Rescue 1122 accident data for 2015-2024. This dataset depicted a gradual upward trend from 2021 onwards coinciding with the Lahore’s systematic SFC rollout. Conclusively, the findings of the current study synthesize directional asymmetry in operational benefits on the corridor and produces overall net disbenefit. The results challenge the popular narrative of traffic operational net benefits on SFCs and highlight the overall safety challenges elevation due to transformation. Present study also provides the first calibrated microsimulation evidence from South Asian mixed-traffic context for this class of urban infrastructure intervention. .

Date Published:2026

UX for AI : A framework for designing ai driven products

Nudelman, Greg

In UX for AI: A Framework for Designing AI-Driven Products, Greg Nudelman ― a seasoned UX designer and AI strategist ― delivers a proven framework for creating exciting, effective, and practical UX designs that integrate with new artificial intelligence technologies. You’ll find tools you can put to work immediately as you research, plan, design, and test user experiences that incorporate human-AI interactions.The book offers hands-on design techniques that rely on tested user research methods. It also includes exercises, like Digital Twin and Value Matrix, that build your design skills and confidence, as well as AI patterns and best practices that have proven effective in real-world deployments of LLMs (Large Language Models), AI agents, copilots, AI search tools, and more.UX for AI is an original and up-to-date design guide that belongs on the desk of every UX designer and product professional working with AI-enabled or AI-driven products. It’s also a must-read for students of UX design, new and trainee UX professionals, and anyone else involved in the creation of software that incorporates artificial intelligence.Unlock the future of design with UX for AI―your indispensable guide to not only surviving but thriving in a world powered by artificial intelligence. Whether you're a seasoned UX designer or a budding design student, this book offers a lifeline for navigating the new normal, ensuring you stay relevant, valuable, and indispensable to your organization.In UX for AI: A Framework for Designing AI-Driven Products, Greg Nudelman―a seasoned UX designer and AI strategist―delivers a battle-tested framework that helps you keep your edge, thrive in your design job, and seize the opportunities AI brings to the table. Drawing on insights from 35 real-world AI projects and acknowledging the hard truth that 85% of AI initiatives fail, this book equips you with the practical skills you need to reverse those odds.You'll gain powerful tools to research, plan, design, and test user experiences that seamlessly integrate human-AI interactions. From practical design techniques to proven user research methods, this is the essential guide for anyone determined to create AI products that not only succeed but set new standards of value and impact.

Date Published:2025

Timeframe

Abid, Zuna [22L-7873]

Submitted in fulfillment of the requirements for the degree of BS in Software Engineering to the Department of Software Engineering. ListofFigures ListofTables 1 Introduction 1.1 PurposeofthisDocument 1.3 Definitions,Acronyms,andAbbreviations 1.3.1 Abbreviations 1.3.2 Definitions: 1.4 Conclusion 2 ProjectVision 2.1 ProblemDomainOverview. 2.2 ProblemStatement 2.3 ProblemElaboration 2.3.1 Real-TimeProcessingRequirements 2.3.2 VisualRealismandAnimation 2.3.3 PersonalityConsistency 2.3.4 ResourceEfficiency 2.3.5 ContextualUnderstanding 2.3.6 EmotionalEngagement 2.4 GoalsandObjectives 2.4.1 CreateLifelikeFace-to-FaceDigitalExperiences 2.4.2 BringHistoricalFigurestoLifeThroughConversationalAI 2.4.3 OfferAccessibleVirtualConsultation 2.4.4 MakeConversationsPersonalandEmotionallyConnected 2.4.5 DevelopEfficientSystemArchitecture 2.4.6 ImplementModular,ExtensibleDesign TABLEOFCONTENTS 2.5 ProjectScope 2.5.1 CoreSystemDevelopment 2.5.2 UserInterfaceDevelopment 2.5.3 CharacterDevelopment 2.5.4 BackendInfrastructure 2.5.5 TestingandEvaluation 2.5.6 CompletetechnicalDocumentation 2.5.7 ExplicitlyOutofScope 2.6 SustainableDevelopmentGoal(SDG) 2.7 Constraints 2.7.1 TechnicalConstraints 2.7.2 ResourceConstraints 2.7.3 APIandServiceConstraints 2.7.4 ScopeConstraints 2.7.5 EthicalandLegalConstraints 2.8 BusinessOpportunity 2.8.1 MarketAnalysis 2.8.2 RevenueModels 2.8.3 CompetitiveAdvantages 2.8.4 MarketEntryStrategy 2.9 StakeholdersDescription/UserCharacteristics 2.9.1 StakeholdersSummary 2.9.2 KeyHigh-LevelGoalsandProblemsofStakeholders 2.10Conclusion 3 LiteratureReview/RelatedWork 3.1 Definitions,Acronyms,andAbbreviations 3.2 DetailedLiteratureReview 3.2.1 HeyGen 3.2.2 Disney’sTalkingTurtle 3.2.3 Character.AI 3.3 RelatedWorkSummaryTable 3.4 Conclusion TABLEOFCONTENTS 4 SoftwareRequirementSpecifications 4.1 ListofFeatures 4.2 FunctionalRequirements 4.3 QualityAttributes 4.4 Non-FunctionalRequirements 4.4.1 Performance 4.4.2 Scalability 4.4.3 Availability 4.4.4 Portability 4.4.5 Security 4.4.6 Accessibility 4.4.7 Usability 4.5 Assumptions 4.6 UseCases 4.7 HardwareandSoftwareRequirements 4.8 GraphicalUserInterface 4.9 DatabaseDesign 4.10 RiskAnalysis 4.11Conclusion 5 High-LevelandLow-LevelDesign 5.1 SystemOverview 5.2 DesignConsiderations 5.2.2 GeneralConstraints 5.2.3 GoalsandGuidelines 5.2.4 DevelopmentMethods 5.3 SystemArchitecture. 5.3.1 ClientLayer 5.3.2 ServerLayer 5.3.3 ExternalServicesLayer 5.3.4 SubsystemArchitecture 5.4 ArchitecturalStrategies 5.4.1 CommunicationMechanisms 5.4.2 ReuseofExistingSoftwareComponents TABLEOFCONTENTS 5.4.3 FuturePlans 5.5 DomainModel/ClassDiagram 5.6 PoliciesandTactics 5.6.1 Policy/Tactic1:ErrorHandlingandRecovery. 5.6.2 Policy/Tactic2:PerformanceOptimization 5.6.3 Policy/Tactic3:SecurityandPrivacy 5.6.4 Policy/Tactic4:TestingandQualityAssurance 5.7 Conclusion 6 ImplementationandTestCases 6.1 Implementation 6.1.1 Speech-to-Text(STT)Implementation 6.1.2 LLMImplementation 6.1.3 VideoGenerationandStreaming 6.2 TestcaseDesignanddescription 6.3 TestMetrics 6.4 Conclusion 7 UserManual 7.1 LaunchingtheApplication 7.2 CreatingaTalkingAvatar(TextBased 7.3 CreatingaTalkingAvatar(ImageUploadMethod) 7.4 StartingConversationwithAvatar 7.5 Real-TimeSpeechInteraction 7.6 ChangingAvatarPersonality 8 ConclusionandFutureWork 8.1 Conclusion 8.2 Futurework 8.2.1 ExpansionintoAdvancedConversationalIntelligence 8.2.2 Multi-PlatformDeployment 8.2.3 Domain-SpecificAIAvatars . Timeframe represents a paradigm shift in human-computer interaction by creating truly immersive con versational experiences with virtual personalities. This project addresses the growing need for accessi ble, personalized digital interactions in education, professional consultation, and entertainment domains. The system integrates multiple advanced technologies including real-time speech recognition, text-to speech synthesis, facial animation with lip-synchronization, and presence detection. These components work seamlessly to deliver natural, emotionally engaging conversations that feel authentic and memo rable. Key innovations include personality-driven response systems that adapt to user context, expressive vi sual animations synchronized with dialogue, and efficient resource management enabling deployment on standard consumer hardware. The platform serves multiple stakeholder groups: educators seeking in teractive learning tools, professionals requiring virtual consultation services, and general users desiring novel entertainment experiences. Initial development has established the core system architecture, evaluated technological frameworks, and developed prototypes demonstrating feasibility. The project scope encompasses complete system development including user interface design, backend integration, character persona development, and comprehensive testing protocols. This report provides detailed documentation of problem analysis, literature review, requirements specifi cation, architectural design, implementation strategies, and future development roadmap for FYP-2. The work completed in FYP-1 validates technical feasibility and establishes a solid foundation for creating a production-ready system that will transform how people interact with AI-powered virtual personalities. .

Date Published:2026

Proactive threat detection: machine learning for suspicious activity recognition

Nasir, Maryam [21L-5685]

Submitted in fulfillment of the requirements for the degree of BS in Data Science to the Department of Data Science. TableofContents ListofFigures ListofTables 1 Introduction 1.1 PurposeofthisDocument 1.2 IntendedAudience 1.3 Definitions,Acronyms,andAbbreviations 1.4 Conclusion 2 ProjectVision 2.1 ProblemDomainOverview 2.2 ProblemStatement 2.3 ProblemElaboration 2.4 GoalsandObjectives 2.5 ProjectScope 2.5.1 ProjectVision. 2.5.2 Exploratory/ResearchPhase 2.5.3 ImplementationPhase 2.6 SustainableDevelopmentGoal(SDG) 2.6.1 SDG16:Peace,Justice,andStrongInstitutions 2.6.2 SDG11:SustainableCitiesandCommunities 2.7 Constraints 2.8 BusinessOpportunity 2.9 StakeholdersDescription/UserCharacteristics. 2.10Conclusion 3 LiteratureReview/RelatedWork 3.1 Definitions,Acronyms,andAbbreviations Page 9 of 86 - AI Writing Submission Submission ID trn:oid:::1:3225938208 Page 9 of 86 - AI Writing Submission Submission ID trn:oid:::1:3225938208 vii Page 10 of 86 - AI Writing Submission TABLE OF CONTENTS ::1:3225938208 3.2 Detailed Literature Review . 3.2.1 M-LLM Based Video Frame Selection for Efficient Video Understanding 3.2.2 MiniGPT4-Video: Advancing Multimodal LLMs for Video Understanding with Interleaved Visual-Textual Tokens 3.2.3 Crowd BehaviorDetection: LeveragingVideoSwinTransformerforCrowdSize and Violence Level Analysis 3.2.4 The use of convolutional neural networks for abnormal behavior recognition in crowd scenes 3.2.5 Propounding First Artificial Intelligence Approach for Predicting Robbery Be havior Potential in an Indoor Security Camera 3.2.6 Object Detection Using Convolutional Neural Networks and Transformer-Based Models: A Review 3.2.7 Suspicious Action Detection in Intelligent Surveillance System Using Action Attribute Modelling 3.2.8 Real-Time Crime Detection Using Convolutional LSTM and YOLOv7 3.2.9 Advanced CCTV Surveillance AnomalyDetection, Alert Generation and Crowd Management using Deep Learning Algorithm 3.2.10 Reviewing Approaches and Techniques for Detecting Suspicious Human Be havior: A Comprehensive Survey 3.2.11 Suspicious Behavior Detection with Temporal Feature Extraction and Time Series Classification for Shoplifting Crime Prevention 3.2.12 Fasttrack: Ahighly efficient and generic gpu-based multi-object tracking method with parallel kalman filter 3.2.13 FPGA implementation of multi-dimensional Kalman filter for object tracking and motion detection 3.2.14 Visual object tracking: A survey 3.2.15 A motion model based on recurrent neural networks for visual object tracking 3.2.16 Human abnormal behavior detection using CNNs in crowded and uncrowded surveillance- A survey . 3.2.17 A novel approach for suspicious activity detection with deep learning 3.2.18 Suspicious Actions Detection System Using Enhanced CNN and Surveillance Video 3.2.19 Suspicious activities detection using spatial–temporal features based on vision transformer and recurrent neural network Page 10 of 86 - AI Writing Submission Submission ID trn:oid:::1:3225938208 TABLEOFCONTENTS 3.2.20 V3Trans-Crowd: AVideo-basedVisualTransformer forCrowdManagement Monitoring 3.3 LiteratureReviewSummaryTable 3.4 Conclusion 4 SoftwareRequirementSpecifications 4.1 ListofFeatures 4.2 FunctionalRequirements 4.3 QualityAttributes 4.3.1 Reliability. 4.3.2 Maintainability 4.3.3 Usability 4.3.4 Correctness 4.4 Non-FunctionalRequirements 4.4.1 Availability 4.4.2 Reusability 4.4.3 Robustness 4.4.4 SecurityRequirements 4.4.5 Performance 4.5 Assumptions 4.6 UseCases 4.6.1 SignUp 4.6.2 Login 4.6.4 ViewDetectionResults 4.6.5 GenerateReport 4.6.6 LogHistory 4.6.7 DeleteHistory 4.6.8 SignOut 4.7 HardwareandSoftwareRequirements 4.7.1 HardwareRequirements 4.7.2 SoftwareRequirements 4.8 GraphicalUserInterface 4.9 DatabaseDesign 4.9.1 ERDiagram . Page 11 of 86 - AI Writing Submission Submission ID trn:oid:::1:3225938208 Page 11 of 86 - AI Writing Submission Submission ID trn:oid:::1:3225938208 TABLEOFCONTENTS 4.9.2 DataDictionary. 4.10 RiskAnalysis 4.10.1 TechnicalRisks 4.10.2 BusinessRisks 4.10.3 EthicalandPrivacyRisks 4.11Conclusion 5 ProposedApproachandMethodology 5.1 VideoPreprocessing 5.2 ObjectDetection 5.3 ObjectTracking . 5.4 BehavioralAnalysis 5.5 AlertGeneration 5.6 Conclusion 6 High-LevelandLow-LevelDesign 6.1 SystemOverview 6.1.1 VideoPreprocessing 6.1.2 ObjectDetection 6.1.3 ObjectTracking 6.1.4 AlertGenerationandSecurityDashboard 6.2 DesignConsiderations 6.2.1 AssumptionsandDependencies 6.2.2 GeneralConstraints 6.2.3 GoalsandGuidelines 6.2.4 DevelopmentMethods 6.3 SystemArchitecture. 6.3.1 SubsystemArchitecture 6.4 ArchitecturalStrategies 6.4.1 IntegrationwithExternalAPIs 6.4.2 FutureExtensibility 6.4.3 ErrorDetectionandRecovery 6.4.4 PerformanceOptimization 6.5 DomainModel/ClassDiagram 6.6 PoliciesandTactics . Page 12 of 86 - AI Writing Submission Submission ID trn:oid:::1:3225938208 Page 12 of 86 - AI Writing Submission Submission ID trn:oid:::1:3225938208 TABLEOFCONTENTS 6.6.1 Tools 6.6.2 UserPolicyandSecurity 6.6.3 MaintenanceandImprovement 6.6.4 TestingtheSystem 6.7 Conclusion 7 ImplementationandTestCases 7.1 Implementation 7.1.1 YOLOv7ModelforObjectDetection 7.1.2 DeepSORTforObjectTracking 7.1.3 LSTM-BasedBehavioralAnalysis 7.1.4 AlertSystemandReporting 7.2 Conclusion 8 ExperimentalResultsandDiscussion 8.2 ExperimentalSetup 8.3 PerformanceMetrics 8.4 Results 8.5 ComparativeAnalysis 8.6 VisualResults 8.7 Discussion 8.8 Conclusion 9 ConclusionandFutureWork 9.1 Conclusion 9.2 FutureWork. As complex threats are evolving day by day, the need for proactive surveillance and threat detection system has never been more critical. Traditional systems rely heavily on human monitoring, leading to delays and errors. To address these challenges, we propose a system that leverage the power of computer vision and machine learning to perform video analytics and detection of suspicious activities in surveillance videos. The purpose of this document, as discussed in the first chapter, is to explore the automation of suspicious activity recognition using machine learning techniques. The project’s goals are outlined in the second chapter. These include developing an advanced object detection system, utilizing behavioral analysis models, and integrating an intelligent response mechanism. By achieving these goals, the project aims to provide a scalable solution for efficient security monitoring. The document proceeds to delve into an extensive literature review in the third chapter, providing com prehensive insights into previous research conducted within our problem domain. Key component high lighted by the literary works, which could lead us to a potential solution, include effectiveness of deep learning models in anomaly detection. The fourth chapter discusses the functional and non-functional requirements, along with other software and hardware prerequisites necessary for the development of our system. Additionally, the chapter provides an in-depth analysis of use cases as well as the graphical user interface (GUI). Finally, potential risks associated with this project are also examined. The proposed approach and methodology are outlined in the fifth chapter. The proposed system involves three key modules: video preprocessing, object detection, and behavioral analysis, each designed to contribute to the accurate recognition of threats. This chapter provides a detailed overview of each module, including its functionality and employed algorithms. The sixth chapter provides an overview of the proposed system, its high-level and low-level designs, and the interactions between various system components. These elements lay the groundwork for the development and implementation of ThreatVision, aiming to deliver a robust and user-friendly system. The seventh chapter marks a significant milestone in the project’s progression toward a comprehensive solution. It details the full implementation of the ThreatVision system including its object detection, object tracking, behavioral analysis, alert system and reporting modules. The chapter illustrates the transformation of the theoretical framework into an operational system. The eighth chapter provides comprehensive experimental results for the ThreatVision system. This Page 7 of 86 - AI Writing Submission Submission ID trn:oid:::1:3225938208 Page 8 of 86 - AI Writing Submission Submission ID trn:oid:::1:3225938208 chapter highlights the experimental setup, performance metrics and the results, including the compara tive analysis of difference surveillance systems, traditionalm motion-based and our proposed AI-based system.

Date Published:2026

Machine learning Animated

Liu, Mark

Section I Installing Python and Learning Animations 1. Installing Anaconda and Jupyter Notebook 2. Creating Animations Section II Machine Learning Basics 3. Machine Learning: An Overview 4. Gradient Descent - Where the Magic Happens 5. Introduction to Neural Networks 6. Activation Functions Section III Binary and Multi-Category Classifications 7. Binary Classifications 8. Convolutional Neural Networks 9. Multi-Category Image Classifications Section IV Developing Deep Learning Game Strategies 10. Deep Learning Game Strategies 11. Deep Learning in the Cart Pole Game 12. Deep Learning in Multi-Player Games 13. Deep Learning in Connect Four Section V Reinforcement Learning 14. Introduction to Reinforcement Learning 15. Q-Learning with Continuous States 16. Solving Real-World Problems with Machine Learning Section VI Deep Reinforcement Learning 17. Deep Q-Learning 18. Policy-Based Deep Reinforcement Learning 19. The Policy Gradient Method in Breakout 20. Double Deep Q-Learning 21. Space Invaders with Double Deep Q-Learning 22. Scaling Up Double Deep Q-Learning. The release of ChatGPT has kicked off an arms race in Machine Learning (ML), however ML has also been described as a black box and very hard to understand. Machine Learning, Animated eases you into basic ML concepts and summarizes the learning process in three words: initialize, adjust and repeat. This is illustrated step by step with animation to show how machines learn: from initial parameter values to adjusting each step, to the final converged parameters and predictions.This book teaches readers to create their own neural networks with dense and convolutional layers, and use them to make binary and multi-category classifications. Readers will learn how to build deep learning game strategies and combine this with reinforcement learning, witnessing AI achieve super-human performance in Atari games such as Breakout, Space Invaders, Seaquest and Beam Rider.Written in a clear and concise style, illustrated with animations and images, this book is particularly appealing to readers with no background in computer science, mathematics or statistics.

Date Published:2024

Security in NDN

Mobeen, Muhammad [24L-7715]

Submitted in fulfillment of the requirements for the degree of MS in Computer Science to the Department of Computer Science . Acknowledgements Author’s Declaration Certificate of Approval Abstract Table of Contents List of Figures List of Tables 1 Introduction 1.1 Architecture for NDN–IoT 1.2 State of the Art 1.3 Motivation 1.4 Problem Statement 1.5 Research Questions 2 Literature Review 2.1 Lightweight Security Mechanisms for IoT and Their Limitations 2.1.1 Proxy Re-Encryption in NDN 2.1.2 Edge-Assisted Security and Trust-Based Approaches 2.2 Summary of Related Work 2.2.1 Surveys and Foundational Architecture 2.2.2 Proxy Re-Encryption, Lightweight Cryptography, and Access Control 2.2.3 Privacy, Attack Detection, and Resilience 2.2.4 Wireless NDN, Caching, and Migration 2.2.5 Gap Identified 2.2.6 Proxy Re-Encryption Background 3 Proposed Framework and Methodology 3.1 System Model 3.1.1 Entities 3.2 Threat Model 3.3 Dynamic Behavioral Trust Mechanism 3.3.1 Comparison of Trust and Reputation Mechanisms 3.3.2 Behavioural Signals 3.3.3 Trust Score and EMA Update 3.3.4 Trust Levels and Policy 3.4 The Five SLICERT-NDN Algorithms 3.4.1 Algorithm 1: Registration 3.4.2 Algorithm 2: Lightweight Interest 3.4.3 Algorithm 3: Edge Trust Evaluation and ECC Processing 3.4.4 Algorithm 4: Secure Delivery 3.4.5 Algorithm 5: Producer Content Encryption 4 Experimental Results 4.1 Implementation Environment 4.1.1 Network Topology 4.1.2 Evaluation Results 4.2 IoT Cryptographic Cost and Scalability 4.2.1 IoT Cryptographic Cost 4.2.2 Scalability with Device Count 4.3 Network Performance 4.3.1 Interest Satisfaction Ratio 4.3.2 Communication Overhead 4.3.3 Edge Cache Hit Rate 4.4 Trust Mechanism Evaluation 4.4.1 Population and Trust Trajectories 4.4.2 Blacklist 4.4.3 Detection Accuracy 5 Conclusion and Future Work 5.1 Summary of Contributions 5.2 Future Work Referencee . The convergence of Internet of Things (IoT) and Named Data Networking (NDN) promises efficient delivery of content through Intra-network caching and data-centric security. However, applying standard NDN security on resource-constrained IoT hardware introduces heavy processing, which existing schemes such as PRIDN (Privacy Preserving Data sharing in NDN) assume the device can run. On IoT devices, these operations cost several seconds per request, which is impractical for battery-powered nodes. This thesis proposes Secure Lightweight IoT Communication with Edge Re-Encryption and Trust in Named Data Networking SLICERT-NDN, a lightweight security framework that combines Proxy Re-Encryption (PRE) with a runtime behavioral trust mechanism, without modifying the standard NDN certificate format. A per-device trust score is updated continuously using an Exponential Moving Average (EMA) and the devices that fall below a threshold are added to a blacklist. The framework is evaluated using ndnSIM on a wireless WiFi 802.11 IoT topology with a wired backbone, and is verified using Verifpal under an active Dolev-Yao adversary. The results show that the cryptographic cost per-cycle of IoT drops by 96.9%, the IoT energy is roughly 1.2x, the Interest satisfaction Ratio (ISR) achieves the ideal satisfaction rate of 1.0%, and the trust mechanism achieves an overall accuracy of 0.991%.

Date Published:2026

Practical Machine Learning : A beginner's guide with ethical insights

Nyamawe, Ally S.

1. Fundamentals of Machine Learning- 2. Mathematics for Machine Learning- 3. Data Preparation- 4. Machine Learning Operations- 5. Machine Learning Software and Hardware Requirements- 6. Responsible AI and Explainable AI- 7. Artificial General Intelligence- 8. Machine Learning Step-by-Step Practical Examples. . The book provides an accessible, comprehensive introduction for beginners to machine learning, equipping them with the fundamental skills and techniques essential for this field.It enables beginners to construct practical, real-world solutions powered by machine learning across diverse application domains. It demonstrates the fundamental techniques involved in data collection, integration, cleansing, transformation, development, and deployment of machine learning models. This book emphasizes the importance of integrating responsible and explainable AI into machine learning models, ensuring these principles are prioritized rather than treated as an afterthought. To support learning, this book also offers information on accessing additional machine learning resources such as datasets, libraries, pre-trained models, and tools for tracking machine learning models. This is a core resource for students and instructors of machine learning and data science looking for a beginner-friendly material which offers real-world applications and takes ethical discussions into account.

Date Published:2025

AI for Communication

Gunkel, David J.

1. Introduction: AI and Communication 2. Machine Translation 3. Natural Language Processing 4. Social Robots 5. Large Language Models 6. Computational Creativity and Generative AI 7. The Future of Communication. AI for Communication offers an engaging exploration into the diverse applications of artificial intelligence (AI) within the realm of communication. By bridging the gap between the scientific and engineering realms of AI and communication, this book reveals how AI, since its inception during the Dartmouth Summer workshop of 1956, has inherently been a science of communication. Exploring key advancements such as machine translation, natural language processing, large language models, computational creativity, and social robotics, this book shows how these innovations not only disrupt but also actively transform human communication.The book is designed for students, teachers, and general readers who want to know how the field of communication impacts and influences the theory and practice of AI and how recent developments in AI will affect all aspects of human social interaction.

Date Published:2025

Agentic clinical precision framework (CPF) for med-VQA

Hanif, Muhammad Nouman [24K-8001]

Submitted in fulfillment of the requirements for the degree of MS in Data Science to the Department of Computer Science. 1 Introduction 1.1 Background and Motivation 1.2 Problem Statement 1.3 Research Objectives 1.4 Research Questions 1.5 Overview of the Proposed Solution 1.6 Scope and Interpretation of Safety 1.7 Key Contributions 1.8 Thesis Organization 2 Literature Review 2.1 The Evolution of Med-VQA Models: From Classification to Generation 2.1.1 Early Approaches (Classification-Based) 2.1.2 The Paradigm Shift to Generative Models (VLMs) 2.2 The Importance of Domain-Specific Pre-training 2.3 The Critical Challenge of Evaluation in Med-VQA 2.3.1 Insufficiency of Standard NLP Metrics 2.3.2 The Need for Faithfulness, Interpretability, and Trust 2.3.3 Safety, Automation Bias, and Governance 2.3.4 Shortcut Learning and the “Black Box” Problem 2.3.5 Emerging Solutions in VQA Evaluation 2.3.6 Mitigation-Oriented Alternatives 2.3.7 The System 2 Reasoning Gap in Evaluated Models 2.4 Intervention Landscape Across the Med-VQA Pipeline 2.5 Identifying the Research Gap: The Motivation for the CPF 3 Methodology and Experimental Setup 3.1 Theoretical Foundation: Desiderata for Evaluation 3.2 Overall Research Framework 3.3 Phase 1: Data Curation and Experimental Baseline 3.3.1 Composite Dataset Engineering 3.3.2 The Multi-Stage Pipeline 3.3.3 Baseline Model Selection 3.4 Phase 2: The Clinical Precision Framework (CPF) 3.4.1 Experimental Induction of the Fluency Trap (Onco-VQA) 3.4.2 The Multi-Agent LLM-as-a-Judge Paradigm 3.4.3 The Multi-Agent Tribunal Architecture 3.4.4 Evaluation Dimensions and Scoring Rubric 3.4.5 Mathematical Arbitration: The Deterministic Veto 3.5 CPF-Guided Self-Correction Strategy 3.5.1 Architectural Challenge 3.5.2 Execution Pipeline 3.6 Implementation Strategy 3.6.1 Experimental Setup 3.6.2 Validation Protocol 4 Results and Analysis 4.1 Phase 1: Baseline Evaluation and The Evaluation Gap 4.1.1 ZS1: Zero-Shot General Evaluation 4.1.2 ZS2: Zero-Shot Abnormality Evaluation 4.1.3 FT: Fine-Tuning Evaluation 4.1.4 ZS2 vs. FT: The Impact of Fine-Tuning 4.1.5 Qualitative Analysis and Discussion 4.2 Phase 2: CPF Evaluation from Oncology to Global Audit 4.2.1 The “Fluency Trap” in Oncology (Onco-VQA Results) 4.2.2 Global Safety Profile at Scale 4.3 CPF-Guided Self-Correction Results 4.3.1 Empirical Recovery Metrics 4.3.2 Interpretation 4.3.3 Metric Correlation Analysis 4.3.4 Taxonomy of Medical VLM Failures 4.4 Summary of Findings 5 Conclusion and Future Work 5.1 Conclusion 5.2 Limitations 5.3 Directions for Future Research 5.3.1 From Post-Hoc Evaluation to Active Intervention 5.3.2 Architectural Mitigation of Multimodal Sycophancy 5.3.3 Reinforcement Learning from Clinical Feedback (RLCF) 5.3.4 Real-World Clinical Deployment and Validation 5.4 Final Remarks References . The integration of Artificial Intelligence (AI) into clinical diagnostics has accelerated interest in Medical Visual Question Answering (Med-VQA), where models must interpret radiological images and answer natural-language questions about findings, anatomy, and pathology. Despite rapid progress in generative Vision-Language Models (VLMs), stronger linguistic fluency does not reliably correspond to stronger clinical safety. Current evaluation protocols remain dominated by general-domain Natural Language Processing (NLP) metrics such as BLEU, ROUGE, and BERTScore, which emphasize lexical and semantic overlap rather than diagnostic validity. As a result, these metrics fall into the “Fluency Trap”: they can assign strong scores to outputs that sound medically plausible while still containing clinically dangerous hallucinations. This thesis addresses that evaluation gap through a two-phase study. First, an empirical baseline was established by merging VQA-RAD [1], VQA-Med 2019 [2], and SLAKE [3] into a composite benchmark of more than 24,000 question-answer pairs and evaluating multiple Med-VQA models under zero-shot and fine-tuned settings. The results confirmed that fine-tuning often improves medical fluency while simultaneously increasing hallucination risk. Second, the Clinical Precision Framework (CPF) is introduced as a deterministic, multi-agent evaluation architecture for safety-aware auditing. The CPF uses specialized agents for diagnostic accuracy, clinical utility, hallucination detection, visual grounding, and report synthesis, and scores predictions across six clinical dimensions using a safety-first arbitration rule (CPS = V × Sbase). Applied to a concentrated oncology benchmark and a 20,284-sample global audit, the framework exposed failure patterns that conventional metrics masked, including high rates of clinically unsafe yet semantically fluent responses. The study further includes judge-alignment and anchoring-gap analyses, identifies multimodal sycophancy under visual verification, and demonstrates a CPF-guided self-correction loop that recovers the NLP-metric (more aligned to the reference strings) in second-pass. Overall, the thesis shows that clinically grounded auditing requires more than lexical similarity and that CPF provides a scalable evaluation framework to support safer benchmarking and analysis of generative medical AI. .

Date Published:2026

Startup in a box

Abdullah, Muhammad [22L-6712]

Submitted in fulfillment of the requirements for the degree of BS in Computer Science to the Department of Computer Science. ListofFigures ListofTables 1 Introduction 1.1 PurposeofthisDocument 1.2 IntendedAudience 1.3 Definitions,Acronyms,andAbbreviations 1.4 Conclusion 2 ProjectVision 2.1 ProblemDomainOverview 2.2 ProblemStatement 2.3 ProblemElaboration 2.3.1 EstablishingaDigitalPresence 2.3.2 ContentandMarketing 2.3.3 MarketAwareness 2.3.4 InvestmentPreparation 2.4 GoalsandObjectives 2.5 ProjectScope 2.6 SustainableDevelopmentGoal(SDG) 2.7 Constraints 2.8 BusinessOpportunity 2.9 StakeholdersDescription/UserCharacteristics 2.9.1 StakeholdersSummary 2.9.2 KeyHigh-LevelGoalsandProblemsofStakeholders 2.10Conclusion 3 LiteratureReview/RelatedWork 3.1 Definitions,Acronyms,andAbbreviations 3.2 DetailedLiteratureReview 3.2.1 WixADI 3.2.2 JasperAI(formerlyJarvis) 3.2.3 SEMrush 3.2.4 Beautiful.ai 3.2.5 AutoGPT 3.2.6 HubSpotCMSandMarketingHub 3.2.7 Looka(formerlyLogojoy) 3.2.8 Webflow 3.2.9 CoScheduleMarketingSuite 3.2.10 Crayon 3.2.11 Copy.ai 3.2.12 Gamma 3.2.13 LivePlan 3.2.14 Hootsuite 3.2.15 Unbounce 3.3 LiteratureReviewSummaryTable 3.4 Conclusion 4 SoftwareRequirementSpecifications 4.1 ListofFeatures 4.2 FunctionalRequirements 4.2.1 WebsiteGeneration 4.2.2 MarketResearch 4.2.3 Pitchdeck 4.2.4 MarketingContent 4.2.5 ProgressTracking 4.3 QualityAttributes 4.4 Non-FunctionalRequirements 4.5 Assumptions 4.6 UseCases 4.7 HardwareandSoftwareRequirements 4.7.1 HardwareRequirements 4.7.2 SoftwareRequirements 4.8 GraphicalUserInterface 4.9 DatabaseDesign 4.9.1 DataDictionary 4.10RiskAnalysis 4.10.1 TechnicalRisks 4.10.2 ProgrammaticRisks 4.10.3 OperationalRisks 4.11Conclusion 5 High-LevelandLow-LevelDesign 5.1 SystemOverview 5.2 DesignConsiderations 5.2.1 AssumptionsandDependencies 5.2.2 GeneralConstraints. 5.2.3 GoalsandGuidelines 5.2.4 DevelopmentMethods 5.3 SystemArchitecture 5.3.1 FrontendArchitecture 5.3.2 BackendArchitecture. 5.3.3 DatabaseArchitecture 5.4 ArchitecturalStrategies 5.4.1 AgentOrchestrationStrategy 5.4.2 PromptEngineeringStrategy 5.4.3 ErrorHandlingandRecoveryStrategy. 5.5 DomainModel/ClassDiagram 5.6 PoliciesandTactics 5.6.1 Tools 5.6.2 CodingGuidelines 5.6.3 TestingStrategy 5.6.4 MaintainabilityandExtensibility 5.6.5 Conclusion 6 ImplementationandTestCases 6.1 Implementation 6.1.1 FrontendImplementation 6.1.2 BackendImplementation 6.1.3 DatabaseImplementation 6.2 TestCaseDesignandDescription 6.2.1 UserAuthenticationTestCases 6.2.2 BusinessProfileManagementTestCases 6.2.3 WebsiteGenerationTestCases 6.2.4 MarketingContentGenerationTestCases 6.2.5 MarketAnalysisTestCases 6.2.6 PitchDeckGenerationTestCases 6.2.7 DownloadTestCases 6.3 TestMetrics 6.4 Conclusion 7 UserManual 7.1 GettingStarted 7.2 CreatingaBusinessProfile 7.3 GeneratingaWebsite 7.4 GeneratingMarketingContent 7.5 GenerateaMarketAnalysis 7.6 GeneratingaPitchDeck 8 ConclusionandWorkCompleted. Startup in a Box solves several challenges that aspiring entrepreneurs and small business owners face when taking a business online. Suck as a fully functional website, some social media presence, market analysis to get information about the business, and a pitch if you want to attract investors. All of this requires a lot of time, money, and coding skills which new entrepreneurs don’t have. Entrepreneurs nowadays must negotiate a disjointed array of specialized resources and services. It could take weeks to learn how to utilize a website builder or hire a developer to create a website. Design and authoring expertise are essential when producing marketing materials. Having access to pricey market research tools and being able to translate data into strategy are essential for comprehending the com petitive landscape. Creating a pitch is difficult as it requires a understanding of what the investors want and how you can convince them. Small startups lack the resource needed for such tasks as they need different skill sets and platforms. Our advanced multi-agent AI system will transform these fragmented tools into an intelligent and seam less experience. By just writing their business idea, users can meet all the requirements to start an online business. This system will create a website, social media content such as videos as posts, perform mar ket analysis, and make a pitch deck. By doing so will act as a virtual co-founder and a full startup team. In addition to utilizing specialized AI agents that operate together under an intelligent orchestrator ar chitecture, the system is based on a robust and modern technology stack that consists of MongoDB for flexible data storage, Node.js with Express for the backend, and React for the frontend. While the website agent understands modern online design concepts, the pitch deck agent arranges narratives for investor appeal, the competitor agent studies market dynamics, and the marketing agent produces en gaging content. Each agent specializes in different tasks, but they all work together to ensure that every module and output is relevant to the business.

Date Published:2026

ReFrame

Iqbal, M. Nouman [22L-6671]

Submitted in fulfillment of the requirements for the degree of BS in Computer Science to the Department of Computer Science. ListofFigures ListofTables 1 Introduction 1.1 PurposeofthisDocument 1.2 IntendedAudience 1.3 Definitions,Acronyms,andAbbreviations 1.4 Conclusion 2 ProjectVision 2.1 ProblemDomainOverview 2.2 ProblemStatement 2.3 ProblemElaboration 2.4 GoalsandObjectives 2.5 ProjectScope 2.5.1 InScope 2.5.2 Deliverables 2.5.3 OutofScope 2.6 SustainableDevelopmentGoal(SDG) 2.7 Constraints 2.8 BusinessOpportunity 2.9 StakeholdersDescription/UserCharacteristics. 2.9.1 StakeholdersSummary 2.9.2 KeyHigh-LevelGoalsandProblemsofStakeholders 2.10Conclusion 3 LiteratureReview/RelatedWork 3.1 Definitions,Acronyms,andAbbreviations 3.2 DetailedLiteratureReview 3.2.1 AI-PoweredEducationalAnimationPlatforms 3.2.2 LargeLanguageModelsinSTEMEducation 3.2.3 AIVideoGenerationTechnologies 3.2.4 MathematicalReasoningandVisualUnderstanding 3.2.5 AI-DrivenEducationalPlatformsandTools 3.3 LiteratureReviewSummaryTable 3.4 Conclusion 4 SoftwareRequirementSpecifications 4.1 ListofFeatures 4.2 FunctionalRequirements 4.3 QualityAttributes 4.4 Non-FunctionalRequirements 4.5 Assumptions 4.6 UseCases 4.7 HardwareandSoftwareRequirements 4.7.1 HardwareRequirements 4.7.2 SoftwareRequirements 4.8 GraphicalUserInterface 4.8.1 AuthenticationScreens 4.8.2 DashboardandProjectManagement 4.8.3 ProjectWorkspace 4.9 DatabaseDesign 4.9.1 ERDiagramDescription 4.9.2 DataDictionary 4.10RiskAnalysis 4.11Conclusion 5 High-LevelandLow-LevelDesign 5.1 SystemOverview 5.2 DesignConsiderations 5.2.1 AssumptionsandDependencies 5.2.2 GeneralConstraints 5.2.3 GoalsandGuidelines 5.2.4 DevelopmentMethods 5.3 SystemArchitecture 5.3.1 SubsystemArchitecture:TheProductionAgent 5.4 ArchitecturalStrategies 5.4.1 Real-TimeStreamingviaWebSocket 5.4.2 AgenticWorkflowwithTool-AugmentedLLMsviaMCP 5.5 DomainModel/ClassDiagram 5.6 PoliciesandTactics 5.6.1 ContainerizationforEnvironmentConsistency 5.6.2 StructuredLoggingforObservability 5.7 Conclusion 6 ImplementationandTestCases 6.1 ToolsandTechnologies 6.2 ImplementationDetails 6.2.1 FrontendImplementation 6.2.2 BackendAPI 6.2.3 ProductionAgent–ReActStreamingLoop 6.2.4 MCPToolsServer 6.2.5 RAGPipeline–DocumentationandSkills 6.2.6 Cross-SessionMemoryandExperienceStore 6.2.7 KokoroTTSIntegration 6.2.8 SceneAnalyserandVisualiser 6.2.9 DockerRenderingPipeline 6.3 TestCaseDesignandDescription 6.3.1 TestCase1:BatchScriptEdit–ValidEditsAccepted 6.3.2 TestCase2:BatchScriptEdit–InvalidItemRejected 6.3.3 TestCase3:SceneVisualiser–FrameSelectionPreservesFirstandLast 6.3.4 TestCase4:SceneVisualiser–VisualModelImageInjection 6.3.5 TestCase5:SceneVisualiser–TextFallbackforNon-VisionModels 6.3.6 TestCase6:SessionRuntimeAssets–SVGandSkillsSetup 6.3.7 TestCase7:SVGLibrary–NameandAliasSearch 6.3.8 TestCase8:DocsRAG–MarkdownSectionSplitting 6.4 TestMetrics 6.4.1 IssuesIdentifiedDuringTesting 6.5 Conclusion 7 UserManual 7.1 SystemRequirements 7.2 GettingStarted 7.2.1 CreatinganAccount 7.2.2 SigningIn 7.2.3 ChangingtheTheme 7.3 NavigatingtheDashboard 7.4 CreatingaNewProject 7.5 WorkingintheProjectWorkspace 7.5.1 ChatPanel(Left) 7.5.2 ScenesPanel(Right) 7.5.3 SourcesPanel 7.6 ExportingandSharingVideos 7.7 Troubleshooting 8 Conclusion 8.1 SummaryofWork 8.2 ObjectivesAchieved 8.3 ChallengesFaced 8.4 FinalRemarks. ReFrame addresses a persistent challenge in higher education: translating expert knowledge into com pelling, technically accurate instructional videos without requiring instructors to master complex ani mation software or programming workflows. Educators routinely rely on static slides or generic stock footage because existing creative suites are time-intensive and poorly aligned with STEM (Science, technology, engineering, and mathematics) pedagogy. Our project reimagines this pipeline by pairing large language models with the Manim animation engine to automatically script, visualize, and narrate lesson content that mirrors the clarity of hand-crafted explainer videos. The solution is anchored around four capabilities. First, a multimodal ingestion layer parses textual prompts, PDF documents, and ref erence imagery to infer the pedagogical objective. Second, an AI planning component decomposes the concept into an animation storyboard that aligns with instructional design principles such as progres sive disclosure and cognitive load management. Third, ReFrame synthesizes Manim code, ensuring mathematical accuracy, consistent visual grammar, and parameterized scenes for iterative refinement. Finally, an integrated text-to-speech pipeline generates high-quality narration that is synchronized with rendered motion, yielding an export-ready video segment. Delivering this experience requires a cohe sive system architecture. The backend orchestrates tool-augmented reasoning where the AI agent calls specialized services, such as code interpreters, PDF extractors, and voice synthesis modules, through secure APIs. Render jobs run in a containerized environment optimized for reproducible Manim out puts, while a metadata layer tracks assets for versioning and reuse. The forthcoming user interface exposes this pipeline through an academic-friendly workflow: instructors upload materials, review au togenerated scripts, trigger renders, and download finalized lessons without touching code. Our devel opment roadmap emphasizes reliability and educational value. Initial milestones focus on core STEM domains such as calculus, linear algebra, and physics, where precise diagrams and stepwise logic are critical. User studies with faculty will validate that ReFrame’s outputs improve learner comprehension and shorten preparation time.

Date Published:2026

Tahaffuz360

Rukh, Mah [22L-6702]

Submitted in fulfillment of the requirements for the degree of BS in Computer Science to the Department of Computer Science. ListofFigures ListofTables 1 Introduction 1.1 PurposeofthisDocument 1.2 IntendedAudience 1.3 Definitions,Acronyms,andAbbreviations 1.3.1 Definitions 1.3.2 Acronyms,andAbbreviations 1.4 Conclusion 2 ProjectVision 2.1 ProblemDomainOverview 2.2 ProblemStatement 2.3 ProblemElaboration 2.4 GoalsandObjectives 2.5 ProjectScope 2.6 SustainableDevelopmentGoal(SDG) 2.7 Constraints 2.8 BusinessOpportunity 2.9 StakeholdersDescription/UserCharacteristics 2.9.1 StakeholdersSummary 2.9.2 KeyHigh-LevelGoalsandProblemsofStakeholders 2.10Conclusion 3 LiteratureReview/RelatedWork 3.1 Definitions,Acronyms,andAbbreviations 3.2 DetailedLiteratureReview 3.2.1 CrimeHotspotPredictionUsingDeepGraphConvolutionalNetworks 3.2.2 DotsversusDensity:Theimpactofcrimemappingtechniquesonperceptionof safety,policeperformanceandneighbourhoodquality 3.2.3 A2DClusteringBasedHotspot IdentificationApproachforSpatio-Temporal CrimePrediction 3.2.4 Aroutediscoverymechanismusing improved tortoisebeetleoptimizer for drivers 3.2.5 AnImplementationofaCrime-Safety-MapApplicationBasedonaSafetyIndex 3.2.6 SecureRatePlanningandSafeRoutingviaCrimeHotspotDetection 3.2.7 AccidentAnalysisandPrevention 3.2.8 SafeRoute–LearningtoNavigateStreetsSafelyinanUrbanEnvironment 3.2.9 TheSafe:ARobustModelforSafestRoutePredictionusingCrimeandAcci dentalData 3.2.10 PCSAPublicSafetyApp 3.2.11 SafetyMap 3.2.12 PunjabPolice–WomenSafetyApp 3.2.13 Life360 3.2.14 GeoSure 3.3 LiteratureReviewSummaryTable 3.3.1 SummaryofResearch-BasedStudies 3.3.2 SummaryofDevelopment-BasedApplications 3.4 Conclusion 4 SoftwareRequirementSpecifications 4.1 ListofFeatures 4.2 FunctionalRequirements 4.2.1 FunctionalRequirementsforUsers 4.3 QualityAttributes 4.3.1 Usability 4.3.2 Reliability 4.3.3 Performance 4.3.4 Scalability 4.3.5 Maintainability 4.3.6 Security 4.4 Non-FunctionalRequirements 4.5 Assumptions 4.6 UseCases 4.6.1 UserRegistration 4.6.2 ReportanEmergency. 4.6.3 SOSAlert 4.6.4 ViewCrimeHeatmap 4.6.5 SafestRouteGeneration 4.6.6 ViewRiskReports 4.6.7 UpdateEmergencyContacts 4.6.8 ViewSafetyGuidelines. 4.6.9 SendLocation 4.6.10 ViewProfile. 4.7 HardwareandSoftwareRequirements 4.7.1 HardwareRequirements 4.7.2 SoftwareRequirements 4.8 GraphicalUserInterface(GUI) 4.8.1 RegisterandMapPages 4.8.2 ReportIncidentPage 4.8.3 FeedandGuidelinesPages 4.8.4 ProfilePage 4.9 DatabaseDesign 4.9.1 Entity–RelationshipDiagram(ERD) 4.9.2 DataDictionary 4.10RiskAnalysis 4.11Conclusion 5 ProposedApproachandMethodology 5.1 Overview 5.2 DataAcquisitionandPreprocessing 5.2.1 DataSources 5.2.2 DataFeatures 5.2.3 DataCleaningandNormalization 5.3 WeightedRisk-BasedRouteScoringSystem 5.3.1 Methodology 5.3.2 RiskWeightCalibration 5.3.3 ContinuousLearningandUpdating 5.4 Real-TimeEmergencyReportingandSOSAlerts 5.4.1 SOSActivation 5.4.2 EmergencyResponseHandling 5.5 DynamicHeatmapGeneration 5.5.1 Time-BasedHeatZones 5.5.2 VisualizationMethod 5.6 AdminVerificationandUserCredibilitySystem 5.7 Conclusion 6 High-LevelandLow-LevelDesign 6.1 SystemOverview 6.2 DesignConsiderations 6.2.1 Assumptions 6.2.2 Dependencies 6.2.3 Constraints 6.3 SystemArchitecture 6.4 Module-LevelDesign 6.4.1 EmergencyReportingModule 6.4.2 RouteScoringModule 6.4.3 HeatmapModule 6.5 ClassDiagram 6.6 TestingandValidationPlan 6.6.1 FunctionalTesting 6.6.2 PerformanceTesting 6.6.3 RouteAccuracyEvaluation 6.6.4 UsabilityTesting 6.7 Conclusion 7 ImplementationandTestCases 7.1 Implementation 7.1.1 ImplementationofDataPreparationandFeatureEngineering 7.1.2 ImplementationofAHP-BasedRiskScoringAlgorithm 7.1.3 ImplementationofHeatmapandSpatialAggregation 7.1.4 ImplementationofRisk-AwareRouteComputation 7.1.5 ImplementationofReal-TimeReportingandEmergencySystem 7.2 TestCaseDesignandDescription 7.2.1 CommonTestConditions 7.2.2 DetailedTestCases 7.3 TestMetrics 7.4 TraceabilityMatrix 7.5 Conclusion 8 UserManual 8.1 Introduction 8.2 SystemRequirements 8.3 InstallationGuide 8.4 ApplicationWorkflow 8.4.1 SplashScreen 8.4.2 Onboarding 8.4.3 UserRegistrationandLogin 8.5 HomeScreen(Dashboard) 8.6 SafetyMapUsage 8.7 IncidentReporting 8.7.1 QuickReport 8.7.2 DetailedReport 8.8 RoutePlanning 8.9 FeedScreen 8.10EmergencyMode 8.11 ProfileManagement 8.12 Settings 8.13Troubleshooting 8.14Conclusion 9 ExperimentalResultsandDiscussion 9.1 Introduction 9.2 DatasetOverview 9.3 AHP-BasedRiskScoringResults 9.3.1 RiskScoreDistribution 9.3.2 TemporalAnalysis 9.3.3 SpatialAnalysis 9.4 HeatmapEvaluation 9.4.1 VisualizationResults 9.4.2 TemporalAdaptation 9.4.3 Observations 9.5 RouteOptimizationResults 9.5.1 RouteComparison 9.5.2 PerformanceAnalysis 9.5.3 AlgorithmBehavior 9.6 System-LevelEvaluation 9.6.1 Real-TimeReportingImpact 9.6.2 EmergencySystemTesting 9.7 Discussion 9.8 Conclusion 10Conclusions 10.1Conclusion 10.2 FutureWork 10.2.1 IntegrationofReal-WorldData. 10.2.2 MachineLearning-BasedRiskPrediction 10.2.3 AdvancedRoutingOptimization 10.2.4 EnhancedVerificationandModerationSystem 10.2.5 Cross-PlatformExpansion 10.2.6 UserExperienceEnhancements 10.2.7 ScalabilitytoOtherCities 10.3 FinalRemarks. Tahaffuz360 is aimed to fill a critical gap in the contemporary urban navigation i.e, the absence of safety-oriented route planning. Although the available applications are efficient in terms of time and distance optimization, they do not consider the most important aspect, which is personal security, and leave commuters (especially women and susceptible communities) without the idea of the risks existing in their travelling directions. This project sees a future in which all people will be able to travel freely as it will create an intelligent and community-driven application on smartphones that will enable its users to have real-time safety intelligence and proactive navigation features. The core of Tahaffuz360 is a complex route scoring system which determines the safety of the possible paths. This is done by incorporating and synthesizing varied data such as past open data of U.S. cities on crimes and accidents, current and user-reported incidences, and contextual details such as time of the day and distance to safe areas. This information is fed in a weighted risk algorithm to produce a dynamic Safety Index of the route segments. It then uses the algorithm of Dijkstra to determine and provide to the user the safest possible route, which is somewhat a priority to security over efficiency.

Date Published:2026

Compartive analysis of collocational errors of ESL learners of a private and a public sector university in Pakistan

Ahmed, Mohammad Maqsood [24L-8628]

Submitted in fulfillment of the requirements for the degree of MS in Science & Humanities . Abstract LIST OF TABLES LIST OF FIGURES Abbreviations 1 INTRODUCTION 1.1 Background of the study 1.2 Problem Statement 1.3 Research Objectives 1.4 Research Questions 1.5 Purpose of the study 1.6 Theoretical framework 1.6.1 Collocational Classification Framework 1.6.2 Thematic Analysis 1.7 Significance of the research 1.8 Delimitation of the study 2 LITERATURE REVIEW 2.1 Describing Collocation 2.1.1 Lexical Collocation 2.1.2 Grammatical Collocations 2.2 Collocational Errors 2.3 Methodological Implications of Collocations 2.4 Contrastive Analysis 2.5 Error Analysis 3 RESEARCH METHODOLOGY 3.1 Study Design 3.2 Sample 3.3 Theoretical Framework 3.4 Data Collection Tools 3.5 Data Analysis 3.6 Ethical Considerations 4 ANALYSIS AND RESULTS 4.1 INTRODUCTION 4.2 Descriptive Statistical Analysis 4.3 Qualitative Analysis 4.3.1 Teachers’ Understanding of Collocation 4.3.2 Classroom Practices for Teaching Collocation 4.3.3 Assessment Practices and Teaching Challenges 4.3.3.1 Impact of Limited Authentic Exposure 5 DISCUSSION AND CONCLUSION 5.1 INTRODUCTION 5.2 Discussion 5.3 Conclusion 5.4 Research Limitations 5.5 Implications of the Study 5.6 Recommendations for Research References Appendices Appendix-A Appendix B . The study conducted Comparative Analysis of Collocational Errors of ESL Learners of a Private and a Public Sector University in Pakistan. The researcher utilised a mixed method approach for the current study. Quantitative analysis was followed by qualitative analysis. The study adopted an explanatory sequential design. A number of 180 descriptive essays were collected from a public and a private university, each group with 90 essays. In addition, the qualitative data was collected through taking interviews from eight ELT teachers. The interview consisted of seven questions which intended to identify the potential causes of committing collocational errors. The quantitative analysis was carried by identifying and comparing errors, equating through descriptive statistics and interpreting the data. In addition, the qualitative analysis took place by extracting emerging themes and subthemes from the transcribed data of the interviews and elaborating them critically. The key findings revealed that both groups committed massive errors in Adverb + Adjective category. After that, the second highest errors were recorded in Verb + Noun variable. The third highest error committed in Verb + Adverb category. More importantly, Noun + Noun appeared to be the most stable variable before Adjective + Noun. Moreover, the study revealed that limited exposure, direct translation and L1 influence were major causes of committing collocational errors. The research suggested that learners must be exposed to authentic materials, contextualised learning, chunk vocabulary learning, such as formulaic sequences. The study concluded that even advanced ESL learners perceived difficulties in collocational usage. The key challenges were negative transfer, direct translation and overgeneralization.

Date Published:2026

Data-driven learning-based model predictive control for output power regulation of electro-surgical instruments

Javaid, Arsal [24L-8502]

Submitted in fulfillment of the requirements for the degree of MS in Electrical Engineering to the Department of Electrical Engineering. . Table of Contents Chapter 1 – Introduction 1.1 Background and Context 1.2 Electrosurgical Power Regulation: Current State and Challenges 1.3 Problem Statement 1.4 Research Objectives 1.5 Research Contributions 1.6 Scope and Limitations 1.7 Thesis Organization Chapter 2 – System Description 2.1 Electrosurgical Technology and Clinical Requirements 2.2 Power Converter Topologies and Modeling 2.3 Control Strategies for DC-DC Converters 2.4 State Estimation 2.5 Data-Driven and Learning-Based Control 2.6 Hardware-in-the-Loop Testing and Gaps Chapter 3 – Mathematical Modeling 3.1 Overall System Architecture 3.2 Buck Converter Modeling 3.2.1 Switch-ON Mode (0 ≤ t < DTsw) 3.2.2 Switch-OFF Mode (DTsw ≤ t < Tsw) 3.2.3 State-Space Averaged Model 3.3 Exact Discretization via Matrix Exponential 3.4 Full-Bridge Inverter Model 3.6 Simulink Implementation Chapter 4 – Baseline MPC Design 4.1 MPC Formulation 4.2 Extended Kalman Filter 4.3 Three-Event Gain Scheduler 4.4 Integral Action, Anti-Windup, and Soft-Start 4.5 Complete Control Pipeline Chapter 5 – Data-Driven MPC Variants 5.1 ANFIS-Based Integral MPC 5.2 Reinforcement Learning-Based Integral MPC 5.3 Fuzzy Logic Integral MPC with Hardware-in-the-Loop Chapter 6 – Results and Discussion 6.1 Simulation Test Scenarios 6.1.1 Test 1: Setpoint Tracking 6.1.2 Test 2: Step Load Disturbance 6.1.3 Test 3: Switching Load Disturbance 6.1.4 Test 4: Input Voltage Variation 6.2 Performance Metrics 6.3 PID Controller Results 6.4 Pure MPC (No Integral Action) Results 6.5 RL Based Integral MPC Results 6.6 ANFIS Based Integral MPC Results 6.7 Comparative Analysis of Simulation Based Controllers 6.7.1 Quantitative Comparison Tables 6.7.2 Discussion of Comparative Results 6.8 Hardware in the Loop Validation: Fuzzy Integral MPC on ESP32 6.9 Extended Kalman Filter Validation 6.10 Summary of Key Findings Chapter 7 – Conclusion and Future Work 7.1 Summary of Contributions 7.2 Key Findings 7.3 Recommendations for Future Work References. . Electrosurgical units (ESUs) must regulate output power across tissue impedances varying from 50 Ω to over 3000 Ω within milliseconds. Conventional controllers respond too slowly, resulting in thermal overshoot and increased collateral tissue damage. This work presents a data-driven, learning-based Model Predictive Control (MPC) framework for robust ESU power regulation in a buck-converter-fed full-bridge inverter. A baseline MPC incorporating Extended Kalman Filter (EKF) state estimation, phase-plane gain scheduling, integral action, and conditional anti-windup is extended into three adaptive variants that automatically tune the integral gain using different learning paradigms: An Adaptive Neuro-Fuzzy Inference System (ANFIS), a Twin Delayed Deep Deterministic Policy Gradient (TD3) reinforcement learning agent, and a compact fuzzy controller implemented in hardware-in-the-loop (HIL). Under setpoint-tracking, load-step, and combined-disturbance scenarios, the proposed learning-based controllers substantially outperform conventional PID control, which exhibits 80% overshoot, 44 V steady-state error, and fails to settle within 0.5 s. Compared with PID, baseline MPC reduces the integral squared error (ISE) by 77% but retains steady-state offset due to model mismatch. Among the adaptive approaches, the ANFIS-based MPC achieves the best overall performance with 0.27% overshoot, 0.116 V steady-state error, and an 88.15 ms settling time, making it the preferred solution for safety-critical electrosurgical applications. The TD3-based MPC delivers the lowest error indices (ISE = 59.11, ITAE = 0.0573) owing to its faster transient response but exhibits 1.19% overshoot and 0.256 V steady-state error, while analysis reveals that the S-curve gain scheduler stabilizes reinforcement learning at the cost of limiting policy adaptation. HIL validation on an ESP32 confirms real-time feasibility of the complete control pipeline within a control loop, achieving a worst-case steady-state error of 0.042 V despite a communication-limited 10 kHz control rate. These results demonstrate that learning-enhanced MPC provides fast, accurate and embedded-realizable power regulation for next-generation electrosurgical systems. .

Date Published:2026