Credit Card Fraud Detection Based on a Hybrid CNN-RNN Deep Learning Model

Abstract

Credit card fraud detection is essential for protecting financial systems by promptly identifying unauthorized or anomalous transactions. Leveraging the strengths of Deep Learning (DL), this paper explores multiple architectures, including the Convolutional Neural Network (CNN), Multi-Layer Perceptron (MLP), Deep Neural Network (DNN), Recurrent Neural Network (RNN), and a hybrid CNN-RNN model, for detecting fraudulent behavior within transactional data. Using a balanced dataset of 559,856 records obtained from a publicly available Kaggle repository, each model was evaluated based on accuracy, precision, recall, F1-score, and ROC-AUC. Experimental results showed that the CNN-RNN hybrid model outperformed all other models, achieving 99.99% accuracy, 0.9971 precision, perfect recall (1.0000), 0.9985 F1-score, and a ROC-AUC of 1.0000. These findings highlight the CNN-RNN model's effectiveness in real-time fraud detection, offering exceptional classification performance with minimal false positives and maximum anomaly coverage.