Detection of credit card frauds with machine learning solutions: an experimental approach

Abstract

In many cases frauds in payment transactions could be detected by analysing the customer’s behaviour. Only in the United States fraudulent transactions led to financial losses of 300 billion a year. Machine learning (ML) and Data Mining techniques were shown to be efficient for detection of fraudulent transactions. This paper proposes an experimental way for designing a ML solution to the problem, which allows practitioners to minimise financial losses by analysing the customer’s behaviour and common patterns of using credit cards. The solution designed within a Random Forest (RF) strategy is examined on a public data set available for the research community. The results obtained on the benchmark data show that the proposed approach provides a high accuracy of detecting fraudulent transaction based on the customer’s behaviour patterns that were learnt from data. This allow us to conclude that the use of the RF models for detecting credit card fraud transactions allows practitioners to design an efficient solution in terms of sensitivity and specificity. Our experimental results show that practitioners using the RF models can find new insights into the problem and minimise the losses.