A Comparative Analysis of Machine Learning Models for Financial Fraud Detection

Abstract

Increasing cyber attacks, like fraud detection in financial transactions, were emerging as a new critical challenge in global and Indian markets. The study introduces a new, improved fraud detection framework integrated with a machine learning model. It includes Random Forest and XGBoost. It improves the development of a scalable, adaptable real-time analysis of the dynamic financial dataset. The proposed methodology achieves excellent accuracy in detecting transactions. This will easily distinguish legitimate active transactions from fraudulent ones. It will be capable of identifying from a highly imbalanced dataset. This extensive experiment conducted on a real-world financial dataset demonstrates the system’s features. It will ensure enhanced security measures and minimize false positives and risks. It regains the consumer’s trust in digital financial services. The proposed machine learning technology gives a high detection accuracy of 92%. Better results were also cited compared to the state of the art from other approaches in the key metrics, such as precision, recall, and F1-score. Results identify improvement through such performance graphs and pie charts showing both the performance of false positives and the improved fraud detection accuracy of the system. This research aims to bridge the crucial gaps in current approaches. We are charting a future course to improve the security and techniques of digital payment systems. The future course encompasses the incorporation of block chain technologies. The system approach, along with new emerging technologies, counters cyber threats.