Hybrid Sampling and Variant Deep Learning Models for Credit Card Fraud Detection

Credit card fraud detection poses significant chal-lenges due to transaction data's imbalanced nature and fraud- sters' continuously evolving strategies. This paper proposes an efficient approach to fraud detection by integrating hybrid sampling techniques with advanced deep-learning models. By combining under-sampling and over-sampling methods, we ad-dress data imbalance and enhance the performance of our detection models. Our approach employs various Deep Learning (DL) architectures, including Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), Long Short-Term Mem-ory (LSTM), and Gated Recurrent Units (GRU). Experimental results demonstrate that hybrid sampling techniques, combined with these DL models, significantly improve detection accuracy and reduce false positives, offering a robust solution for real- world credit card fraud detection. Among DL models, GRU outperformed the others, achieving 99.51 % precision, 99.96% recall, 99.73% accuracy, and 99.73% Fl-score on the credit card fraud dataset.