FRAUD DETECTION IN BANKING DATA BY MACHINE LEARNING TECHNIQUES

Abstract

As technology advanced and e-commerce services expanded, credit cards became one 
of the most popular payment methods, resulting in an increase in the volume of 
banking transactions. Furthermore, the significant increase in fraud requires high 
banking transaction costs. As a result, detecting fraudulent activities has become a 
fascinating topic. In this study, we consider the use of class weight-tuning 
hyperparameters to control the weight of fraudulent and legitimate transactions. We 
use Bayesian optimization in particular to optimize the hyperparameters while 
preserving practical issues such as unbalanced data. We propose weight-tuning as a 
pre-process for unbalanced data, as well as CatBoost and XGBoost to improve the 
performance of the LightGBM method by accounting for the voting mechanism. 
Finally, in order to improve performance even further, we use deep learning to fine
tune the hyperparameters, particularly our proposed weight-tuning one. We perform 
some experiments on real-world data to test the proposed methods. To better cover 
unbalanced datasets, we use recall-precision metrics in addition to the standard ROC
AUC. CatBoost, LightGBM, and XGBoost are evaluated separately using a 5-fold 
cross-validation method. Furthermore, the majority voting ensemble learning method 
is used to assess the performance of the combined algorithms. LightGBM and 
XGBoost achieve the best level criteria of ROC-AUC D 0.95, precision 0.79, recall 
0.80, F1 score 0.79, and MCC 0.79, according to the results. By using deep learning 
and the Bayesian optimization method to tune the hyperparameters, we also meet the 
ROC-AUC D 0.94, precision D 0.80, recall D 0.82, F1 score D 0.81, and MCC D 0.81. 
This is a significant improvement over the cutting-edge methods we compared it to.