AI-Augmented Fraud Detection and Cybersecurity Framework for Digital Payments and E-Commerce Platforms

Abstract

Advanced fraud detection techniques are becoming more and more necessary as e-commerce and digital transactions continue to grow.  The intricacy and changing nature of fraudulent actions can make it difficult for traditional rule-based systems to keep up. To tackle this challenge, this research presents a hybrid fraud detection framework that combines several machine learning techniques, including Logistic Regression, XGBoost, a fusion of Autoencoder with XGBoost, and Graph Neural Networks (GNN). The proposed system also integrates behavioral pattern analysis and real-time risk evaluation, enabling it to adapt swiftly to new threats. Comprehensive testing on both standard and real-world datasets demonstrates the strength of this approach. The Autoencoder-XGBoost combination emerged as the top performer, achieving 97.4% accuracy with precision, recall, and F1-score, all at 0.96, and operating with a latency of just 100 milliseconds. The GNN model also delivered strong results, reaching 96.7% accuracy, a precision of 0.95, a recall of 0.94, and an F1-score of 0.945 while maintaining a lower latency of 88 milliseconds. Comparatively, traditional models like Logistic Regression and standalone XGBoost achieved 89.5% and 94.2% accuracy, respectively. These results highlight the improved effectiveness of hybrid approaches in identifying fraud within modern digital ecosystems.