https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4749 2026-04-30T03:11:11Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4803 Title: Enhancing Creditworthiness in Peer-to-Peer Lending Using Human Centered Artificial Intelligence Authors: Wijetunga, W. L. P. M Abstract: Abstract This research addresses significant gaps in the Peer-to-Peer (P2P) lending field, specifically the lack of transparency, effectiveness, human biases and fairness and high false positive rates. To tackle these issues, design science approach and research onion methodology are utilized with data from the Lending Club P2P lending company. The aim of this research is to make the process of creditworthiness in peer-to-peer lending more effective through the application of Human Centered AI. This involves identifying the most accurate Machine Learning (ML) model, determining the most interpretable eXplainable Artificial Intelligence (XAI) model, integrating both models and evaluating their effectiveness in P2P lending with a focus on interpretability and explainability. The Random Forest classifier is found to be the most accurate ML model compared to XGBoost, LLR and Classification Tree. XAI models such as SHAP, LIME and DiCE provide valuable insights into interpretability. SHAP offers global and local interpretations while LIME focuses on localized explanations. DiCE generates counterfactuals for "what-if" scenarios which help determine necessary changes to loan features. Evaluation includes quantitative metrics such as Accuracy, F1 score and AUC-ROC from ML models as well as qualitative components such as interviews and questionnaires to assess the combined ML and XAI model’s effectiveness. The successful integration of an accurate ML model (Random Forest) with state-of-the-art XAI methods contributes to transparent and efficient creditworthiness assessment in P2P lending. Further research should focus on enhancing the ML and XAI framework through longitudinal studies exploring additional XAI methods across multiple P2P lending platforms. This research sets the foundation for future investigations that will advance the integration of ML and XAI in P2P lending while opening avenues for further improvement in creditworthiness assessment methodologies. 2024-05-01T00:00:00Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4802 Title: Malicious Network Activity Detection through Electromagnetic Radiation in Wired Ethernet Authors: Weerasinghe, B.S Abstract: Abstract Traditional network security measures, while effective in fortifying network infrastructure, face limitations in scenarios where physical interception of network traffic is unfeasible, such as in network forensics investigations or within resource-constrained Internet of Things (IoT) environments. This necessitates the development of noninvasive detection techniques capable of discerning malicious network activities without imposing undue burdens on network resources. In recent years, exploration into Electromagnetic Radiation (EMR) analysis has emerged as a promising avenue for non-invasive monitoring and analysis of network traffic. This research investigates the viability of employing Electromagnetic (EM) side-channel analysis (SCA) to detect malicious network activities in wired Ethernet environments. A hardware setup was devised to simulate an attacker and a victim connected via a Cat 6 cable. Three types of Denial of Service (DoS) attacks (DoS HTTP, DoS TCP, DoS UDP) were simulated across the cable, and respective EM traces were captured. Additionally, benign traffic traces were collected during periods of no intentional communication between the two devices. An H-loop antenna connected to a HackRF One software-defined radio (SDR) device was utilized for data collection. These traces were divided into training and testing datasets, with the training set used to train three models: Random Forest Classifier (RFC) with AdaBoost, Multilayer Perceptron (MLP), and Support Vector Machine (SVM). Subsequently, the model was applied to the testing set, achieving a classification accuracy of 99.70% for distinguishing between normal and malicious traces. These findings demonstrate the feasibility of detecting malicious network-based attacks in a non-invasive manner with sufficient reliability. 2024-05-01T00:00:00Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4801 Title: Estimating Story Points in Scrum: Balancing Accuracy and Interpretability with Explainable AI Authors: Walpita Gamage, I.N Abstract: Abstract This study addresses the challenge of accurate Story Points (SP) estimation in agile software development. SP, a unit for measuring development effort, are crucial for project planning and resource allocation. However, manual SP estimation is critical yet a tedious and error-prone process, often causing delays and exceeding project budgets. This highlights the pressing need for automated and accurate SP prediction in the software development industry. This study addresses this need by proposing a novel data preprocessing approach for story point estimation. It involves removing similar user stories, data augmentation and description segmentation. Furthermore, the study contributes 3 new datasets to the public domain specifically designed for story point estimation research. This enhanced data richness and diversity are shown to significantly improve model performance. The study leverages these 3 datasets and the Choetkiertikul dataset to train various traditional and transformer models, including Support Vector Machines (SVM), Random Forest, Recurrent Neural Network (RNN), Bidirectional Encoder Representations from Transformers (BERT), DistilBERT and RoBERTa. Among the traditional approaches, SVM achieved the highest accuracy (46.12%). BERT outperformed other transformer models (44.58%) but fell slightly short of SVM’s performance. To enhance model transparency and interpretability, the study employed Local Interpretable Model-agnostic Explanations (LIME), Shapley Additive explanations (SHAP), and Transformer Interpret libraries. These techniques offer explanations by highlighting keywords influential in the model’s predictions. Additionally, a human-based evaluation involving 7 industry professionals was conducted to assess both model performance and the reliability of the Explainable Artificial Intelligence (XAI) methods. 2024-05-01T00:00:00Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4800 Title: Consensus-Based Cooperative Location Proof for Untrusted Devices Authors: Vishwajith, K.K.W Abstract: Abstract Modern Location-based Services depend on the user’s honesty to provide the benefits. Since the location is determined mainly using the user devices, adversarial actors abuse these services for their intent. To address this problem we propose a location-proof method that’ll help Location-based Services to serve genuine users of the service positioned in a particular position at a given time. The proposed solution verifies user location by nearby communication method giving proofs of location tied to a boundary provided by the transmission channel. With the range of 30 meters to 50 meters per location-proof, We also address the issues created by having this kind of location method like privacy and security. Verified by Automated Validation of Internet Security Protocols and Applications (AVISPA), we prove that attacks on the protocol can be easily mitigated. We show that the proposed solution is cost-effective since it does not need additional infrastructure and modification to the hardware. 2024-05-01T00:00:00Z