Enhanced Model for Cyber Fraud Detection in the Banking System

Source abstract

The rapid digitization of the banking sector has transformed financial services, enhancing customer convenience through mobile and online platforms, but it has also escalated cyber fraud risks, including phishing, identity theft, and transaction fraud. The rule-based fraud detection systems in banks struggle with high false positive rates, limited scalability, and fragmented integration with cybersecurity frameworks, leading to financial losses and eroded customer trust. This study develops a robust, cybersecurity-enhanced fraud detection framework for the banking ecosystem. Implemented in Python with scikit-learn, the framework leverages real-time analytics via Apache Kafka, advanced feature engineering with temporal attributes i.e. CustomerAvgAmount_7D, and FastAPI for seamless integration with Security Information and Event Management Intrusion Detection Systems. Evaluated on a synthetic dataset of 7,966 transactions, the models of this research outperformed other models, achieving a fraud class F1-score of 0.8721, a false positive rate of 0.001 (7 false positives), and a Receiver Operating Characteristics Area Under Curve of 0.9090, and F1-score 0.8367, FPR 0.006 respectively.