Credit Card Fraud Detection Using Hybrid Proximal Policy Optimization and Artificial Bee Colony Optimization with Mutual Learning

Abstract

The surge in e-commerce has intensified credit card fraud, resulting in massive global losses and creating an urgent need for stronger detection systems. This research presents an advanced model for detecting credit card fraud to address challenges often overlooked, such as class imbalance and sensitivity to initial parameter settings. Our model leverages an artificial neural network (ANN) to extract feature vectors necessary for accurate fraud detection. We utilize proximal policy optimization (PPO) to address class imbalance during training of the ANN. PPO improves the treatment of minority classes by assigning higher rewards for correct predictions and more substantial penalties for errors. This approach leads to more balanced learning. Additionally, our model incorporates a mutual learning-based artificial bee colony (ML-ABC) algorithm for efficiently pre-training the parameters of the ANN. Experiments on the Université Libre de Bruxelles credit card dataset show that the proposed approach achieves 90.197% accuracy and an F-measure of 91.287%. It outperforms the best existing method by about 3%. These results highlight the robustness of the model and its potential for real-world e-commerce fraud detection.