An Optimized LightGBM Model for Fraud Detection

Abstract

Abstract The rapid development of e-commerce and the growing popularity of credit cards have made online transactions smooth and convenient. However, large numbers of online transactions are also the targets of online credit card fraud, which aggregate to enormous losses annually. In response to this trend, many machine learning and deep learning methods have been proposed to solve this problem. Unfortunately, most models have been developed on small datasets and require tedious fine-tuning processes. In this paper, a LightGBM-based method for fraud detection is proposed. The dataset used for this study is the IEEE-CIS Fraud Detection dataset provided by Vesta Corporation, which includes over 1 million samples. Experiments have shown that the LightGBM-based method outperforms most classical methods based on Support Vector Machine, XGBoost, or Random Forest. Besides, effective feature engineering methods for feature selection and Bayesian fine-tuning for automatic hyperparameter searching are also proposed.