Behaviour-Driven Fraud Detection in Multi-User ECommerce Transactions Using Process Mining and Machine Learning

Abstract

This project demonstrates a webbased system for detecting fraudulent transactions using machine learning and process mining techniques. The system allows users to upload transaction datasets, process them to analyze behavioral patterns, and apply multiple machine learning algorithms to identify anomalies. Through an interactive interface, normal and fraudulent activities are visualized, revealing patterns that differentiate legitimate user behavior from potential attacks. The platform supports training on existing datasets and testing on new data to predict fraud in realtime. Performance evaluation of different algorithms indicates that the Extension Random Forest algorithm achieves the highest accuracy. The combination of process mining for behavioral insights and machine learning for predictive analytics provides a robust approach to transaction fraud detection, enhancing security, reducing financial losses, and offering actionable insights for system administrators and analysts.