AutoEncoder and LightGBM for Credit Card Fraud Detection Problems

Abstract

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