A Comparative Evaluation of Deep Learning and Ensemble Algorithms for Online Payment Fraud Detection

Abstract

The exponential growth of digital payment platforms has introduced unprecedented security challenges in detecting fraudulent transactions. This study presents a comprehensive comparative evaluation of deep learning architectures and ensemble learning algorithms for online payment fraud detection. We systematically assess Long Short-Term Memory networks, Recurrent Neural Networks, logistic regression, and gradient boosting methods across detection accuracy, precision-recall trade-offs, and computational efficiency. Through rigorous experimentation on real-world transaction datasets, we evaluate two feature engineering strategies: user behavior-based features from RFM analysis and transaction amount patterns. Our analysis reveals that ensemble methods achieve superior F1-scores of 0.876, while LSTM architectures demonstrate enhanced capability in capturing temporal dependencies. The study establishes quantitative guidelines for algorithm selection based on dataset characteristics and operational constraints.