Using Isolation Forest in anomaly detection: the case of credit card transactions

Abstract

With the evolution of new technology especially in the domain of e-commerce and online banking, the payment by credit card has seen a significant increase. The credit card has become the most used tool for online shopping. This high rate in use brings about fraud and a considerable damage. It is very important to stop fraudulent transactions because they cause huge financial losses over time. The detection of fraudulent transactions is an important application in anomaly detection. There are different approaches to detecting anomalies namely SVM, logistic regression, decision tree and so on. However, they remain limited since they are supervised algorithms that require to be trained by labels in order to know whether the transactions are fraudulent or not. The goal of this paper is to have a credit card fraud detection system which is able to detect the highest number of new transactions in real time with high accuracy.  We will also compare, in this paper, different unsupervised techniques for credit card fraud detection namely LOF, one class SVM, K-means and Isolation Forest so as to single out the best approach.