Machine Learning Techniques for Credit Card Risk Prediction: An Experimental Study

Nowadays, with the rapid development of mobile Internet and smartphones, credit card fraud has become more and more serious, causing significant credit and financial damage to cardholders and economics.Many previous studies proposed using machine learning techniques to predict credit card risks.However, it is still unclear what machine learning models are most suitable for credit card risk prediction.In this paper, we perform an experimental study on real credit card transaction data to evaluate existing machine learning techniques on credit card risk prediction.The unique contributions of the paper are two-fold.First, we present a systematic experimental study to compare eight machine learning techniques on credit card risk prediction.These techniques cover both supervised models and deep learning approaches, providing a relatively complete framework for revealing the performance of different machine learning techniques.Second, we use both imbalanced and balanced datasets and employ various evaluation metrics to measure the performance of machine-learning models.The results show that all models perform better on the balanced dataset than on the imbalanced data, and the Logistic Regression model performs best in most experiments.In addition, deep learning models do not perform as well as supervised models.Finally, we present some helpful suggestions and new insights for future research on credit card risk prediction.