A Blockchain-Integrated Machine Learning Framework for Detecting Fraud in E-Commerce Returns

E-Commerce fraudulent returns pose significant challenges, resulting in substantial financial losses and damaging reputations. This study proposes a hybrid framework for fraud detection, combining blockchain-based rule validation with machine learning models to enhance accuracy and scalability. The framework integrates predefined blockchain rules with machine learning predictions using Random Forest and Reinforcement Learning (Deep Q-Networks with experience replay) for a layered approach to fraud detection. Blockchain smart contracts dynamically verify transaction attributes and user behavior, while the machine learning models adapt to evolving fraudulent patterns. Experimental results demonstrate a marked improvement in accuracy, precision, and recall, alongside a reduction in false positives compared to standalone methods. By leveraging customizable blockchain rules aligned with organizational policies and periodically updating machine learning models, the proposed framework offers a scalable, secure, and adaptive solution for mitigating fraudulent returns in e-commerce.