Secure Banking System with AI Fraud Detection

Abstract

This thesis presents the design, implementation, and evaluation of a secure banking system with integrated artificial intelligence for fraud detection. The research addresses the critical challenge of financial fraud in digital banking platforms through a comprehensive approach combining advanced machine learning techniques with robust security architecture. The proposed system employs a microservices architecture to ensure scalability, fault tolerance, and security isolation. At its core, an AI-powered fraud detection service analyzes user behavior patterns and transaction characteristics in real-time to identify potentially fraudulent activities. The system implements enhanced threshold classification techniques that improve upon traditional binary classification methods, resulting in higher precision and recall metrics even with imbalanced datasets. Additionally, the research explores the integration of a risk assessment engine that complements the machine learning model with rule-based analysis. This hybrid approach provides both the adaptability of AI and the explainability of rule-based systems. The implementation leverages Docker containerization to ensure consistent deployment across environments while maintaining security isolation between components. Experimental results demonstrate significant improvements over traditional fraud detection approaches, with the proposed system achieving 93.7% accuracy and 91.2% precision in identifying fraudulent transactions while maintaining a low false positive rate of 3.8%. The thesis contributes to the field of financial cybersecurity by presenting a comprehensive architecture that can be adapted by banking institutions to enhance their fraud prevention capabilities while maintaining high performance and user experience standards.