Examining ensemble models to detect credit card fraudulent transactions

Abstract

Fraudulent credit card transactions impact both consumers and card issuers. The ability to detect fraudulent credit card transactions can reduce the cost of credit card use. Prior research has shown that machine learning and ensemble models can identify fraudulent transactions with good accuracy. However, no study has been found that compares heterogeneous and homogeneous models. This research study examines and compares machine learning algorithms with multiple ensemble models for detecting fraudulent credit card transactions using data available from a U.S.-based credit card issuer. The results show that heterogeneous ensemble models can better detect fraudulent and non-fraudulent transactions than individual and homogeneous models. The results suggest that underlying individual algorithms are used in the ensemble matter. Specifically, heterogeneous models that use both random forest modeling and neural network modeling tend to outperform individual models and ensemble models that do not utilize both.