Enhanced E-Banking Security System for Phishing Detection Using Machine Learning and Privacy Preservation

Source abstract

E-banking systems have revolutionized financial transactions, but they are increasingly vulnerable to phishing attacks that compromise sensitive user data. Phishing detection in e-banking presents several challenges, including the ability to accurately identify deceptive URLs, the presence of imbalanced datasets where phishing attempts are significantly fewer than legitimate activities, and the need for privacy-preserving mechanisms that protect user data during the detection process. Addressing these issues requires the development of advanced machine learning algorithms that can detect phishing attacks with high accuracy while preserving user privacy. This dissertation presents the development of a phishing detection system for e-banking, addressing key challenges such as detecting sophisticated phishing attempts while ensuring data privacy. The system was built using Object-Oriented Design (OOD) methodology, with Python as the programming language. Utilizing the Random Forest Classifier (RFC) and privacy preservation techniques, the model enhances e-banking security. Exploratory Data Analysis (EDA) revealed an imbalanced dataset, which was mitigated through Random OverSampling. Key features were identified using feature importance ranking and correlation matrices, leading to a highly accurate RFC model. To protect sensitive data, differential privacy was incorporated by adding Laplace noise during both training and deployment. The model achieved an accuracy of 98.98%, outperforming other systems, and was deployed via a Flask web application with a user-friendly interface for phishing URL detection. Comparisons with existing models showed the system’s superior performance across key metrics.