Fuzzy Logic-Based WGAN-GP Data Augmentation and Hybrid DL-ML Models for Robust Credit Card Fraud Detection

Abstract

Credit card fraud is one of the most serious threats facing modern financial systems, causing losses estimated in billions of dollars annually worldwide. Traditional fraud detection systems face significant challenges, one of the most critical is the class imbalance problem, where fraudulent transactions account for less than 0.1% of total transactions, making learning and classification processes difficult. This study aims to develop a hybrid system based on deep learning techniques to address this issue by adopting an advanced data generation approach using Wersstein Generative Adversarial Networks Gradient penalty (WGAN-GP), supported by Spectral Normalization and Fuzzy Logic. The proposed system generates realistic synthetic fraudulent transactions, to augment the dataset, addressing class imbalance and improving minority class representation. Several deep learning models were trained on the augmented data, including Convolutional Neural Networks (CNN), Long Short-Term Memory Networks (LSTM), and the XGBoost algorithm. The performance was evaluated using standard metrics such as the F1-score and the Fréchet Inception Distance (FID) to assess the quality of the generated data. Experimental results demonstrated that integrating WGAN-GP with fuzzy logic significantly improves the accuracy and efficiency of fraud detection, proving the effectiveness of the proposed model in handling data imbalance and uncertainty in real-world scenarios.