Integrating Machine Learning and Deep Learning for Accurate Fraud Detection

Credit card fraud detection is a major challenge for the financial industry, and its prevention requires machine learning models to be accurate and efficient in reducing financial losses. This paper proposes a comparative analysis of three most popular machine learning models DT, RF, and XGB that can be applied for fraud detection. To optimize the model’s performance, we apply RFE to select only the most important features for the classification process. Each of the models has been evaluated for their performance through several metrics including accuracy, precision, recall, F1-score, and AUC-ROC, to validate its effectiveness on detecting fraudulent transactions. We have also analyzed each model’s computational efficiency, not only in training but also for inference time because the real-time nature of this task is so crucial. While our results have shown that XGB outperformed DT and RF in both accuracy and AUC-ROC, it really does need fine hyperparameter tuning to optimize its computational cost. The results of the paper provide excellent insights for banks that want to deploy machine learning models for detecting fraud, by offering a compromise between model performance and computational efficiency.