Credit card fraud detection using XGBoost

Abstract

Financial data is some of the most sensitive information stored on the Internet. With the advent of new attack methods in mobile technology, cardholder property is being invaded, making traditional fraud detection systems unable to provide the necessary security. For this reason, we have developed an advanced system that relies on new technologies to immediately detect unusual behavior and prevent inauthentic transactions. Our system is based on machine learning classi?cation algorithms, which are consid-ered the best solution to the aforementioned problem, as they can ?nd sophisticated fraud features that a human simply cannot detect. There are many approaches to detect fraudulent transactions, but not many of them result in high accuracy due to high transaction class imbalance. Tree Boosting has proven to be a highly e?ective and widely used machine learning method for solving various regression and classi?cation problems. In this work, a tree boosting method called Extreme Gradient Boosting Algorithm (XGBoost) is used to address credit card fraud detection and deal with data imbalance. Two XGBoost models are built and their performances are evaluated. Very satisfactory results, related mainly to the accuracy of transaction classi?cation, are obtained for both datasets (balanced/unbalanced).