Detection and Analysis of Credit Card Application Fraud Using Machine Learning Algorithms

Abstract

Fraud is a widespread problem in the financial industry with devastating effects. It is essential to prevent and reduce fraud effectively. Traditional approaches, such as expert system, suffers from the incapability to handle complex problems and tremendous amount of data, while the recent development of various machine learning techniques brings new solutions. With many research works focusing on tackle frauds of credit card transaction or insurance, only few mentioned the identity fraud of credit card application. This article presents a few machine learning models to detect such fraud. We firstly explore and clean up the data. Then 331 expert variables are created with professional consult and selected to 30 to reduce dimensionality of our data. Multiple models, such as logistic regression and decision trees, are built and fit on the training set. Finally, we found that the random forest model performs the best in terms of fraud detection rate, achieving 54% in out-of-time test. The obtained model can be applied in anti-fraud monitoring systems, or a similar model development process can be performed in related business areas to detect fraud and reduce the occurrence of such behaviors.