AutoEncoder and LightGBM for Credit Card Fraud Detection Problems

This paper proposes a method called autoencoder with probabilistic LightGBM (AED-LGB) for detecting credit card frauds. deep learning-based AED-LGB algorithm first extracts low-dimensional feature data from high-dimensional bank using the characteristics of an which has symmetrical network structure, enhancing ability representation learning. The fraud dataset comes real anonymized by and is highly imbalanced, normal far greater than data. For this situation, smote used to resample before putting extracted into LightGBM, making amount non-fraud equal. After comparing resampled non-resampled data, it was found that performance not improved after resampling, concluded more suitable imbalanced Finally, comparable other commonly machine learning algorithms, such as KNN overall improvement 2% in terms ACC index compared KNN. When threshold set 0.2, MCC 4% higher second-highest 30% It shows accuracy, true positive rate, negative Matthew’s correlation coefficient.