Abstract
Credit card fraud is one of the most common forms of fraud in the financial industry. Machine learning and deep learning models have been widely used to predict this fraud, but there are still several challenges to predicting credit card fraud transactions. This study aims to conduct a comprehensive and comparative analysis of modern machine learning, deep learning, hybrid, and transformer models for predicting credit card fraud, with a focus on determining a best-fit model that can effectively and efficiently reduce the financial losses associated with fraud for individuals, businesses, and the entire economy amongst the state of-the art modern techniques. SMOTE and ADASYN oversampling techniques are applied to mitigate class imbalance before the model is trained and their performance evaluated using accuracy, precision, recall, and F1 score metrics. Ensemble methods random forest and XGBoost emerge as top performers, benefiting from operating on diverse decision trees. However, deep learning models like BiLSTM show comparable results, automatically extracting valuable features; this led to the hybridization of the three models.