Advancing Credit Card Fraud Detection with an Optimized Light Gradient Boosting Machine

Abstract

The immense expansion of online sales has made credit cards the major payment method, and at the same time, the number of fraudulent activities has increased in a parallel manner. Fraud detection systems that work well are, therefore, a must-have in the financial world, as they will keep the financial transactions safe and also cut down the economic losses that would otherwise occur. In this paper, an intelligent credit card fraud detection method is suggested that it makes use of an Optimized Light Gradient Boosting Machine (OLightGBM). The OLightGBM model uses Bayesian hyperparameter optimization for LightGBM to learn faster and provide more accurate predictions. The model was tested on two public credit card datasets containing both legitimate and fraudulent transactions from the real world. The performance of the OLightGBM model is compared with that of traditional machine learning algorithms and found to exceed the latter in terms of accuracy, precision, AUC, and F1-score. The findings prove its capability and appropriateness for use in real-time fraud detection systems.