A Hybrid Machine Learning–Driven Fraud Risk Assessment Framework for Secure Online Course Transactions

Abstract

The rapid expansion of online education platforms has led to a significant increase in digital course transactions, along with a rise in chargeback fraud and unauthorized payment disputes. Since online courses are delivered instantly after payment authorization and cannot be revoked, traditional post-transaction fraud detection methods are ineffective and often result in financial losses for education providers. This creates a strong need for proactive fraud prevention mechanisms that assess transaction risk before granting course access.

This project proposes a hybrid machine learning–driven fraud risk assessment framework for secure online course transactions. The system evaluates transactional, behavioral, and contextual features in real time to predict the likelihood of chargeback fraud prior to payment authorization. Logistic Regression is used for interpretable baseline probability estimation, Random Forest captures non-linear behavioral patterns, and Gradient Boosting enhances prediction accuracy by focusing on difficult-to-classify transactions. The outputs of these models are combined to generate a reliable fraud risk score.

Based on the predicted risk level, transactions are classified as low, medium, or high risk and are accordingly approved, verified, or blocked. The proposed framework aims to reduce chargeback incidents, maintain a low false positive rate, and improve transaction security while ensuring a smooth user experience on online education platforms.