AI-BASED FRAUD DETECTION IN E-COMMERCE PAYMENTS: A COMPREHENSIVE RESEARCH STUDY

Abstract

The rise of e-commerce has led to a surge in payment fraud, highlighting the need for improvements in conventional rule-based fraud detection systems that are not adaptive and have high false positive rates. This research introduces a hybrid Artificial Intelligence (AI) approach to detect fraud in e-commerce transactions. It combines supervised machine learning (Random Forest, XGBoost), unsupervised anomaly detection (Isolation Forest, autoencoders) and a Long Short-Term Memory (LSTM) network to model temporal transaction patterns. The model uses advanced feature engineering techniques that include transactional, user behavior, and device information to enhance detection capabilities. The study uses the Synthetic Minority Over-sampling Technique (SMOTE) to overcome class imbalance. The findings show that the hybrid approach improves fraud detection, especially in detecting new types of fraudulent transactions, while achieving a good recall-precision trade-off. The study showcases the power of AI-based approaches in providing scalable, real-time fraud detection and improving security for e-commerce platforms.