A Comparative Analysis of Credit Card Fraud Detection Using Machine Learning and Deep Learning Techniques

Abstract

E-commerce is growing rapidly around the world and this causes a significant increase in credit card transactions, both normal and fraud transactions. Financial institutions throughout the world lose billions because of credit card fraud. Fraudsters have no fixed styles; they always change their behavior and try to learn new technologies that allow them to commit frauds through online transactions. Moreover, they assume that the regular behavior of consumers and fraud patterns change fast. Fraud detection systems have become necessary for banks and financial institutions to minimize their losses. However, not much research has been conducted on credit card fraud detection methodologies, due to the unavailability of credit card transaction dataset for researchers. Also, most of the researchers apply either machine learning or deep learning techniques without comparing the results of applying both techniques in the same dataset. This research aims to apply different machine learning (Logistic Regression, K-Nearest Neighbor, Random Forest) and deep learning (Deep Neural Network, Convolutional Neural Network) techniques in a real-life credit card dataset to choose the most efficient algorithm for detecting fraud transactions. After applying different experiments with different parameters using all the algorithms mentioned before, the Random Forest Algorithm gives slightly better accuracy than Deep Neural Networks.