A Hybrid Deep Learning–XAI Model for Real-time Credit Card Fraud Detection on Imbalanced Transaction Datasets

Abstract

Credit card fraud has increased significantly with the rapid growth of digital payments, posing financial risks to customers and institutions. Traditional rule-based fraud detection systems often fail to identify novel and rapidly evolving fraudulent behaviors due to their static nature. To overcome these restrictions, this study proposes a hybrid deep learning–explainable AI (XAI) framework for real-time credit card fraud detection. This framework utilizes sophisticated model learning and interpretability approaches to address the challenges posed by highly imbalanced transaction datasets effectively. The proposed model integrates a long short-term memory (LSTM)-based anomaly detection module with a gradient boosting classifier for improved fraud identification. To address class imbalance, techniques such as SMOTE, ADASYN, and cost-sensitive learning were applied. The model incorporates SHAP-based explainability to provide transparent interpretations of fraud predictions, increasing trust and accountability. Experiments conducted using the popular Credit Card Fraud dataset and IEEE-CIS Fraud Detection dataset demonstrate that the hybrid model outperformed baseline algorithms in terms of recall, precision, and F1-score. The results confirm that combining deep learning and XAI enhances both performance and interpretability, making this framework suitable for real-time deployment in financial systems.