Optimizing Credit Card Fraud Detection: Random Forest and XGBoost Ensemble

Ever since the development of the credit card transaction system, the number of credit card frauds has been increasing exponentially and this has been a significant and persistent issue that every other organization is trying to deal with but has not found a long-term solution. So, the aim is to develop a better approach to deal with those kinds of fallacious issues. The purpose of this paper is to come up with a better approach while building a model using different machine- learning algorithms which determine whether a transaction completed using a credit card is genuine or not. With the rapid evolution in technology, it has become very difficult to find out the patterns in criminal transactions manually by a group of humans due to the very big data present and significantly increasing every minute. Artificial Intelligence and machine learning give a new opening to solve the above issues as it automate the detection of fraudulent activities and can improve on its own. In this paper, initially, the dataset available on Kaggle [1] is picked up, which then after pre- processing along with different techniques to get better results like SMOTE[2], feature extraction etc. is applied to different machine-learning algorithms found out that random forest and XGBoost give the best accuracy and improve even the other factors like precision, recall etc. when ensembled using voting classifier which showcases exceptional performance, achieving an accuracy of 99.94 percentage.