/library/oar/handle/123456789/134219 2025-12-26T18:24:41Z 2025-12-26T18:24:41Z /library/oar/handle/123456789/141991 2025-12-05T10:17:45Z 2025-01-01T00:00:00Z

Title: AI‐Driven gesture recognition with smart gloves Abstract: This research presents the development of an AI‐driven gesture recognition system aimed to enhance Human‐Computer Interaction through the use of smart gloves. Many emerging applications, such as virtual reality, robotics, and assistive technologies, require detailed motion capture of the hand in three dimensions. Traditional input devices are not designed to capture such motion, whereas wearable solutions like smart gloves offer a practical means of collecting complex motion data for gesture interpretation. This study proposes a system capable of interpreting dynamic hand gestures captured using smart gloves. A custom dataset was collected using Rokoko smart gloves, recording 14 gesture classes from 14 subjects. Time‐series data captured from the smart gloves was preprocessed, and a range of feature extraction methods, including statistical, frequency‐domain, and motion‐based techniques, were applied. Experimental results were carried out to determine which features or combination of features gives the best result. Dimensionality reduction methods, namely Principal Component Analysis and Autoencoders, were examined to optimise the feature space and reduce complexity. A number of classification models were implemented and compared, including Support Vector Machines, K‐Nearest Neighbours, Hidden Markov Models, as well as, deep learning approaches such as CNN‐LSTM networks. Experimental results showed that while most models achieved high accuracy on validation data, up to 93.64%, performance significantly decreased when tested on data from unseen subjects, dropping to 20.39‐28.93%. This highlights the challenge of inter‐subject generalisation. To mitigate this, personalised models were implemented, showing good performance improvements. The SVM classifiers achieved accuracy results ranging from 67.9% to 92.9%, and the majority of precision, recall, and F1 scores exceeding 85%, while CNN‐LSTM models achieved an accuracy above 95% consistently. Precision, recall, and F1‐score values also remained high. This work contributes to the field of gesture recognition by systematically evaluating feature engineering and modelling techniques on multichannel time‐series data. It underscores the importance of personalised learning strategies and provides insight into the practical limitations of real‐world deployment, such as latency and subject variability. Future work may explore domain adaptation, multimodal sensing, and real‐time implementation to further advance robust gesture‐based interfaces. Description: M.Sc.(Melit.)

2025-01-01T00:00:00Z /library/oar/handle/123456789/141983 2025-12-05T09:53:57Z 2025-01-01T00:00:00Z

Title: Table selection using information retrieval techniques for table-agnostic Text-to-SQL Abstract: Text-to-SQL has been effectively addressed using various NLP approaches, enabling the translation of natural language queries into SQL queries. A common prerequisite for these implementations, however, is the availability of the database table during inference. This requirement can pose challenges in scenarios where the table is not readily accessible to users. This work is motivated by the ongoing development of a chatbot tool within a private company, aimed at streamlining database interactions for the users. To address the table accessibility limitation, this study leverages ����������� Retrieval techniques to implement table selection based solely on the natural language query. We finetune pre-trained models like BERT and GIST-NoInstruct using the ColBERT method. We train our models using data we curate in-house by employing established LLM-prompting techniques. We prepare individual training datasets using two negative sampling techniques: uniform distribution and weighted probability distribution. We also experiment with various data fusion techniques such as RRF, CombMNZ, and Linear Combination to combine results from multiple search strategies. Our approach outperforms baseline methods in table retrieval, while also providing a comparative analysis of various retrieval strategies. Description: M.Sc.(Melit.)

2025-01-01T00:00:00Z /library/oar/handle/123456789/141978 2025-12-05T09:48:46Z 2025-01-01T00:00:00Z

Title: EndoAI diagnostics revolutionizing early detection and diagnostics of endometriosis Abstract: Endometriosis is a chronic and debilitating gynaecological disorder affecting approximately 10% of women worldwide. Characterised by the abnormal growth of endometrial‐like tissue outside the uterus, the condition often leads to severe physical pain, emotional distress, and mental health challenges, significantly reducing patients’ quality of life. Despite its high prevalence, diagnosing endometriosis remains a major clinical challenge due to the heterogeneous nature of its symptoms, frequent misdiagnoses, and reliance on invasive procedures to confirm the diagnosis. Consequently, the average diagnostic delay extends up to eight years. This dissertation proposes a four‐stage solution that addresses these challenges by investigating the potential of Artificial Intelligence (AI) techniques to facilitate the early and accurate diagnosis of endometriosis. Specifically, the study develops a multi‐model AI‐driven diagnostic framework that leverages both self‐reported patient symptom data and laparoscopic medical images. Six Machine Learning (ML) algorithms were employed to predict the likelihood of endometriosis based on symptomatology, incorporating feature engineering techniques to optimise model performance. Additionally, eleven Deep Learning (DL) architectures underwent transfer learning to enhance the detection of endometrial lesions from laparoscopic images. The effectiveness of the proposed models was evaluated through a comparative analysis using key performance metrics, such as accuracy, precision, and recall. The results demonstrated that AI‐powered diagnostic tools significantly enhance the identification of endometriosis, with feature selection and hyperparameter tuning playing a crucial role in improving predictive accuracy. This study further identified high‐performing ML and DL models with strong clinical applicability, as well as key symptom‐based features essential for detecting the disease. These findings highlight the transformative potential of AI in medical diagnostics, particularly in addressing the persistent diagnostic delays associated with endometriosis. By integrating AI‐driven methodologies into clinical workflows, healthcare professionals can improve early detection rates, minimise misdiagnoses, and ultimately enhance patient outcomes. Furthermore, this study underscores the feasibility of a self‐diagnostic tool capable of predicting the likelihood of endometriosis, thereby increasing public awareness of the condition and empowering individuals to seek timely medical consultations. This research contributes to the advancement of AI in gynaecological healthcare, offering a pathway toward more efficient, accessible, and reliable diagnostic solutions for endometriosis. Description: M.Sc.(Melit.)

2025-01-01T00:00:00Z /library/oar/handle/123456789/141970 2025-12-05T09:25:18Z 2025-01-01T00:00:00Z

Title: Multimodal fusion for enhanced smart contract reputability analysis in blockchain Abstract: This study explores the limitations of traditional smart contract reputability assessments, which often rely on either static code analysis or isolated transactional data, missing a comprehensive view of a contract’s evolving trustworthiness. Motivated by the need for robust reputability evaluation in blockchain ecosystems, this work introduces a mul‐ timodal data fusion framework to integrate static and dynamic data sources. Existing solutions are effective at anomaly detection and vulnerability analysis but fail to com‐ bine these data types for holistic insights. Our proposed framework employs boosting algorithms with GAN‐based augmentation for opcode embeddings, achieving superior performance in identifying illicit contracts, with a LightGBM model delivering 97.67% accuracy and a recall of 0.942. A CNN‐based autoencoder is incorporated for multi‐ modal anomaly detection, effectively identifying abnormal patterns by leveraging the interplay between static code and transactional behavior. The multimodal integration yielded a 7.25% improvement in recall compared to single‐source models, confirming its enhanced capacity to detect reputability shifts and abnormal behavior. For long‐term monitoring, an LSTM model captures reputability trends, demonstrating low validation loss and minimal prediction lag, ensuring timely and accurate identification of evolving trustworthiness. The results highlight that multimodal fusion significantly enhances pre‐ dictive accuracy, robust anomaly detection, and the ability to model reputability trends, offering a powerful tool for early risk detection and proactive intervention strategies. This research advances decentralized application security by providing a reliable frame‐ work for improving trustworthiness and mitigating potential risks, forming the crux of a sophisticated multimodal data fusion strategy. Description: M.Sc.(Melit.)

2025-01-01T00:00:00Z