http://dspace.dtu.ac.in:8080/jspui/handle/123456789/47 2026-07-01T03:20:04Z http://dspace.dtu.ac.in:8080/jspui/handle/repository/22951 Title: DL-BASED INTRUSION DETECTION FOR IOMT: A COMPARATIVE ANALYSIS OF EMBED-NET, CONV-NET-SVM, AND DEEP-SVM-NET Authors: KUMAR, ASHISH; Kumar, Shailender (SUPERVISOR) Abstract: The swift development of IoMT technology has led to the ability of continuously track ing patients, automating clinical operations, and facilitating interaction between connected devices in contemporary healthcare settings. While these interrelated processes greatly im prove medical efficiency and accessibility, there are major cybersecurity risks present as well. Typically, medical devices have low computing power, run outdated firmware, and lack sophisticated protection systems; therefore, they become easy targets for diverse cy ber attacks such as DoS [7], reconnaissance, spoofing, and data manipulation attacks. It turns out to be hard to cope with dynamic and sophisticated IoMT traffic using common IDS technologies [8], like signature-based IDS and traditional machine learning techniques. The limitations of the existing systems in terms of poor adaptiveness, simplified feature representation, and high levels of false negatives prove the necessity of further research in creating smart IDS techniques. This study introduces a 3-part deep learning solution specifically for an IoMT IDS Embed-Net, Conv-Net-SVM, and Deep-SVM-Net. Each model is a solution to specific se curity problems with IoMT environment. Embed-Net is an approach that embeds device identifiers, flags, and protocol attributes into dense models that are learned from the data and enables these models to represent subtle interactions between heterogeneous network traffic. The Conv-Net-SVM performed convolutional feature extraction followed by a Sup port Vector Machine (SVM) classifier, fusing the advantages of CNN, which is capable of learning the structs in the flow with good margin-based decision boundaries of SVM. Deep SVM-Net is an output layer based on SVM that exploits a novel approach to deep neural net that functions well in separating benign and malicious traffic and simultaneously has a low computational load for resource constrained medical devices. Several IoT-related datasets that included the various cyber-attack patterns were first preprocessed for standardization of scaling, transformations to ordinal encoding, feature selection based on covariance, and SMOTE balancing before training and validation were performed on the datasets, each using a cross validation strategy that ensured stable and generalizable models. All three exhibited excellent intrusion detection results with par ticularly good overall accuracy and low false-negative rates from Embed-Net, outstanding capability in learning attack signatures with high spatial correlations from ConvNetSVM, and superior intrusion detection real-time efficiency with highly reliable binary separation from Deep-SVM-Net. The results collectively point to the superiority of the deep learn ing architecture for improving the security of IoMT, which efficiently adapts to mixed-type features, to an imbalanced dataset, and to medical operational constraints. 2026-06-01T00:00:00Z http://dspace.dtu.ac.in:8080/jspui/handle/repository/22950 Title: EXECUTION-AWARE SELF-REFINING TEXT-TO-SQL GENERATION USING LARGE LANGUAGE MODELS Authors: SHARMA, ANKIT; KUMAR, VINOD ( SUPERVISOR ) Abstract: The increasing need for data-driven systems has resulted in the growth of user-friendly database interaction methods, especially for non-technical users. Natural Language to SQL, also known as Text-to-SQL is hence required for converting natural language queries from users into executable SQL statements, allowing users to fetch data from databases more efficiently. Although, it is a challenging task to generate accurate SQL queries due to the complexity of natural language understanding, schema linking and variety of database structures. Recent progress in Large Language Models (LLMs) have sig nificantly improved the traditional Text-to-SQL methods by enhancing their semantic understanding for generalized SQL generation. This thesis proposes an Execution-Aware Self-Refining Text-to-SQL Generation framework using Large Language Models, where generated SQL queries are refined iteratively through execution feedback and context awareness. This framework integrates prompt engineering, RAG based architectures and execution validation to identify syntax errors, logical disparities and failed query results. The study also presents an overview of benchmark Text-to-SQL datasets such as Spider, WikiSQL, BIRD, and CoSQL representing that execution-aware self-refinement enhances SQL generation performance compared to conventional approaches. Moreover, it high lights challenges related to scalability, explainability and computational efficiency towards making a domain-centric Text-to-SQL systems for a broad range of technical and non technical users. 2026-06-01T00:00:00Z http://dspace.dtu.ac.in:8080/jspui/handle/repository/22947 Title: EXPLAINABLE TEMPORAL TRANSFORMER FOR DISEASE PROGRESSION PREDICTION USING ATTENTION AND SHAP ANALYSIS Authors: CHHIBBER, AJITESH; KUMAR, VINOD ( SUPERVISOR ) Abstract: Parkinson’s disease (PD) is a progressive neurodegenerative disorder characterized by the gradual deterioration of motor functions, significantly affecting the quality of life of patients. Accurate prediction of disease progression is essential for timely clinical inter vention, treatment planning, and personalized patient management. Traditional machine learning approaches often treat clinical observations as independent samples, limiting their ability to capture the temporal dynamics inherent in longitudinal patient data. Al though deep learning models such as Long Short-Term Memory (LSTM) networks have demonstrated improved temporal modeling capabilities, they may struggle to effectively capture long-range dependencies present in disease progression trajectories. This thesis presents an Explainable Feature-Aware Enhanced Temporal Transformer (FAETT) framework for predicting Parkinson’s disease progression using longitudinal voice-based biomarkers from the Parkinson Telemonitoring dataset obtained from the UCI Machine Learning Repository. The proposed framework integrates temporal sequence modeling with self-attention mechanisms to learn complex relationships across historical patient ob servations. A comprehensive preprocessing pipeline involving data cleaning, feature scal ing, temporal sequence generation, and patient-wise train-test partitioning is employed to ensure robust model development and unbiased evaluation. To assess the effectiveness of the proposed approach, FAETT is compared against three baseline models: Random Forest (RF), Long Short-Term Memory (LSTM), and a stan dard Transformer architecture. Model performance is evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R²). Experimental results demonstrate that the proposed FAETT model achieves supe rior predictive performance, attaining an MAE of 1.2688, RMSE of 1.6457, and R² score of 0.9754, outperforming the baseline approaches. The findings indicate that the incor poration of feature-aware temporal attention significantly enhances the model’s ability to capture disease progression patterns. iii To address the interpretability requirements of clinical decision-support systems, SHAP (SHapley Additive exPlanations) analysis is integrated into the framework. The explain ability analysis identifies the most influential vocal biomarkers contributing to disease progression prediction and provides transparent insights into model behavior. Correlation analysis, residual diagnostics, and prediction-performance visualizations further validate the robustness and reliability of the proposed framework. The results demonstrate that the proposed FAETT architecture effectively combines pre dictive accuracy with model interpretability, making it a promising tool for explainable disease progression forecasting in Parkinson’s disease. The study highlights the potential of attention-based temporal learning and explainable artificial intelligence techniques for advancing data-driven healthcare applications. 2026-05-01T00:00:00Z http://dspace.dtu.ac.in:8080/jspui/handle/repository/22944 Title: SOCIAL BIAS IDENTIFICATION AND MITIGATION IN NATURAL LANGUAGE TEXT USING MACHINE LEARNING Authors: KAMBOJ, PRADEEP; KUMAR, SHAILENDER ( SUPERVISOR); GOYAL, VIKRAM (CO - SUPERVISOR ) Abstract: Advanced Artificial Intelligence (AI) methods have enabled the creation of sophisticated large language models (LLMs) capable of generating human-like text and handling a broad spectrum of complex language comprehension tasks. The last decade has seen the advent of LLMs that fill crucial roles across a variety of applications, including automated content generation and summarization, healthcare analytics, legal decision support, conversational agents, and educational technologies. Despite their remarkable abilities, these models often reflect and even amplify the social biases embedded in the large datasets on which they are trained. These biases can manifest as stereotypes or unjust associations related to gender, race, religion, profession, or other social features. When these AI systems are deployed in high-stakes domains where fairness and reliability are paramount, the presence of such biases raises major ethical, social, and technical concerns. As a result, understanding, measuring, and mitigating bias in LLMs has emerged as a prominent research challenge at the forefront of responsible and trustworthy AI. This thesis constitutes a thorough exploration of social bias in natural language text generated by language models (LMs) and LLMs, with a focus on systematic approaches to measuring, evaluating, and mitigating it. The research draws on theoretical, empirical, experimental, and methodological approaches to investigate bias from several angles across the AI pipeline, including word embeddings, contextualized language models, prompt-based inference functions, and fine-tuning strategies. The work focuses on understanding biases across these components and seeks practical solutions to build fairer and more trustworthy generative AI systems. The initial phase of the research investigates gender bias in contextualized word embeddings generated by transformer-based LMs. Word embeddings are the building blocks of language in many NLP systems, and biases encoded in these representations can carry over to downstream applications. The gender direction in the embedding space is extracted, and the gender polarity of profession-related terms (occupation names) with respect to gendered pronouns is calculated, yielding a quantitative framework for measuring one type of bias: that women or men are less likely to pursue certain professions. Indeed, an experimental analysis shows that dynamic embeddings from transformer-based models exhibit substantial gender associations even in the absence of explicit gender information in the input text. To alleviate this problem, we propose a form of post-processing debiasing that modifies the embedding representations to reduce stereotypical associations while preserving the semantic relationships among words. The experimental results show that the proposed method can significantly alleviate gender bias in profession embeddings, thereby balancing the model’s representations. Building on this foundation, the thesis broadens the analysis to large language models and a wider range of societal biases stemming from multiple demographic attributes. We introduce a systematic evaluation framework for bias in LLM-generated outputs, in part by creating a curated inference dataset from previously established bias benchmarks. The dataset includes contexts that encourage language models to generate stereotypical, anti-stereotypical, and neutral responses, enabling systematic assessment of model behaviour. This study provides a comprehensive mechanism for v analyzing how different models respond to socially sensitive contexts and how bias manifests in generated text. This research makes an important contribution by exploring prompt engineering to both detect and mitigate bias in LLMs. Several types of prompt variants are developed to investigate the effects of their design on model behaviour, namely standard, chain of-thought, cognitive-style, and human-persona prompts. These prompts are systematically assessed to study the effects of various prompting techniques on output bias. Also proposed are the debiased versions of these prompts that explicitly elicit neutral reasoning and unbiased decision-making. The introduction of prompt-only bias evaluation is a key aspect of the extended work, exploring whether biased responses can be induced by prompts alone, without context. Experimental results indicate that when certain prompts are presented to language models, those models make stereotypical predictions, suggesting that bias arises from the interaction between prompts and the models' reasoning mechanisms, rather than solely from the training data. This underlined the importance of careful prompt design and evaluation when deploying language models in real-world settings. Alongside this bias analysis, the research also delves into the issue of hallucination in LLMs, whereby a model provides confident answers that are factually incorrect or unsupported. Across most domains, hallucinations undermine the model’s reliability and may introduce risks in critical domains such as healthcare, legal advice, and policy analysis. To tackle this phenomenon, the thesis presents a contrastive decoding method powered by disturb prompts to compare the probability distributions of model outputs for same prompt and perturbation-prompt scenarios. The method helps detect hallucinated content and enhances the factual consistency of outputs by comparing responses to normal prompts with those to perturbed prompts. The results show that contrastive prompting methods can mitigate hallucination and improve the robustness of language model outputs. Another important aspect of the research is assessing how well fine-tuning approaches mitigate biases. Among such models, large open-source language models are fine tuned on balanced sets with equal numbers of biased/unbiased statements across a wide range of social categories. Fine-tuning is when models are trained to produce more neutral and fair responses while retaining their language comprehension. In fact, experimental results show that fine-tuning with fairness-aware special prompts significantly reduces the model's biased outputs and improves fairness performance. In conclusion, the work in this thesis demonstrates that bias in LMs is a complex, multifaceted phenomenon with multiple underlying sources, including training data, representation learning, and prompting. Tackling this challenge requires the integrated use of bias measurement, dataset design, prompt engineering, model fine-tuning, and evaluation metrics. The methodologies are cross-disciplinary, offering actionable tools to identify and prevent bias in generative AI systems without sacrificing performance or usability. This work extends beyond technical contributions, establishing the need for a broader meaning of fair and responsible development in the internalization of AI. Overall, this thesis gives a good overview of bias in LMs and LLMs. The research, by integrating representation-level analysis, prompt-based evaluation, hallucination detection, and fairness-aware fine-tuning, provides novel insights into the mechanisms that produce vi biases in AI systems while suggesting appropriate strategies to mitigate them. The results of this work demonstrate the potential to help establish more ethical, fair, transparent, and socially responsible generative AI technologies that can serve a wider range of communities without perpetuating harmful stereotypes or obesity-related inequalities. 2026-04-01T00:00:00Z