Effects of Feature Selection with Machine Learning Algorithms in Detection of Credit Card Fraud

Abstract

As an effect of developments in e-commerce systems and communication technologies, Credit cards have become the most common mode of payment for purchases. The payments through the credit cards also involve the risk of credit card fraud such as application fraud, identity theft, lost/stolen card misuse, and phishing. These frauds lead to huge losses and require automatic and real-time fraud detection. Many studies have used Machine Learning (ML) techniques to detect fraudulent transactions. This study focuses on proposing a framework for the detection of credit card frauds by applying machine learning techniques like Random Forest (RF) and Naïve Bayes and testing the results on balanced and unbalanced datasets with and without performing the feature selection on the dataset. After comparing the results, it was discovered that Random Forest outperformed the Naïve Bayes on a balanced dataset with feature selection performed using Recursive Feature Elimination and Information Gain.