Evaluates hybrid approaches for improving credit card fraud detection, highlighting effective machine and deep learning methods.
Key Points
The aim is to enhance the accuracy and efficiency of credit card fraud detection using machine learning and deep learning techniques. The study also addresses data imbalance in transaction datasets.
Applied synthetic minority oversampling technique (SMOTE) and SMOTE-ENN hybrid sampling to balance datasets
Evaluated 37 machine learning and deep learning models for fraud detection
Utilized SHAP and LIME for interpretability of model outputs
Developed two stacking ensemble approaches combining various algorithms, including Extra Trees, CNN, LSTM, and XGBoost
Achieved exceptional model performance with accuracy, precision, recall, F1-score, and AUC all reaching 1.0
Stacking ensemble methods showed significant advancements over traditional models
Deep learning methods, including Feedforward Neural Network and Multilayer Perceptron, demonstrated strong capabilities in detecting fraud patterns