ML-Based Online Transaction Fraud Detection

This paper focuses on fraud detection and the measures necessary to fully automate this process, which has become crucial for banks. With fraud on the rise, it poses significant threats and can result in substantial damages to financial institutions. Transaction data presents unique challenges for fraud detection due to the lack of short-term processing capabilities. The primary objective is to conduct a feasibility study on the selected fraud detection methods. Using various models, we aim to test each transaction individually and proceed accordingly.Initially, we define the detection task, outlining the dataset attributes, metric choices, and techniques to manage unbalanced datasets. This analysis helps identify underlying patterns within the dataset, such as how cardholders’ purchasing habits may evolve over time and how fraudsters may adapt their tactics. We then explore various methods used to derive sequential features from credit card transactions. Financial fraud, the practice of gaining financial benefits through deceitful and unlawful means, has become a significant threat to businesses and organizations. Despite numerous efforts to combat financial fraud, it continues to inflict substantial economic and societal harm, with daily losses reaching significant amounts. Traditional methods for fraud detection, introduced years ago, were predominantly manual, making them time-consuming, costly, and prone to errors. Although more research is being conducted, current approaches have not effectively reduced the financial losses associated with fraud. Conventional fraud detection methods, such as manual verifications and inspections, are inefficient, costly, and inaccurate.