Credit Card Fraud Detection with Unbalanced Real and Synthetic dataset using Machine Learning models

Abstract

As credit card becomes one of the most trusted and popular mode of payment for both online as well as regular purchase, cases of fraud associated with it are also rising. It is of utmost importance for financial institutions like banks and credit card companies to find fraudulent credit card transactions in real time so as to withhold any suspicious transaction till they get further confirmation from the customers and ensure customers aren't paying for anything they haven't bought. But the challenge in making such a model that classifies transactions into legitimate and fraudulent transactions is that the legitimate transactions occur much more frequently than the fraud transactions. In this paper we will analyse how the traditional machine learning algorithms handle highly imbalanced data, biased against fraud transactions, and compare it with algorithms that are designed specifically to deal with highly imbalanced data. For fraud detection in mobile money transaction, a few supervised machine learning models (Logistic Regression, Random Forest, Decision Tree, SVM) are tested and compared. All classification models are tested using a synthetic dataset generated by a simulator based on real transaction of a company.