An Experimental Study With Imbalanced Classification Approaches for Credit Card Fraud Detection

Credit card fraud is a criminal offense. It causes severe damage to financial institutions and individuals. Therefore, the detection and prevention of fraudulent activities are critically important to financial institutions. Fraud detection and prevention are costly, time-consuming, and labor-intensive tasks. A number of significant research works have been dedicated to developing innovative solutions to detect different types of fraud. However, these solutions have been proved ineffective. According to Cifa, 33 305 cases of credit card identity fraud were reported between January and June in 2018.(1) Various weaknesses of existing solutions have been reported in the literature. Among them all, the imbalance classification is the most critical and well-known problem. Imbalance classification consists of having a small number of observations of the minority class compared with the majority in the data set. In this problem, the ratio fraud: legitimate is very small, which makes it extremely difficult for the classification algorithm to detect fraud cases. In this paper, we will conduct a rigorous experimental study with the solutions that tackle the imbalance classification problem. We explored these solutions along with the machine learning algorithms used for fraud detection. We identified their weaknesses and summarized the results that we obtained using a credit card fraud labeled dataset. According to this paper, imbalanced classification approaches are ineffective, especially when the data are highly imbalanced. This paper reveals that the existing approaches result in a large number of false alarms, which are costly to financial institutions. This may lead to inaccurate detection as well as increasing the occurrence of fraud cases.