A Comparison of Machine Learning Techniques for Credit Card Fraud Detection

Fraud is a billion-dollar business and it is increasing every year. The PwC global economic crime survey of 2016 suggests that more than one in three (36%) of organizations experienced economic crime [1]. Those results reveal clearly that, despite the millions of dollars being spent to tackle it, economic crime remains a persistent and serious issue.

In the last years, many studies have been performed using data mining to investigate new techniques to detect fraud on the basis of the fraudulent paths [2] and di erent algorithms have been developed to block fraudulent transactions before they are lled. However, new fraud behaviors born every time, above all in the Internet world, and for this reason we need a continuous improvement of those algorithms.

In the next sections, we use a simulated sample of ~ 200k credit card transactions to test two machine learning algorithms for fraud detection: Logistic Regression (LR) and Random Forest (RF) [3]. We also compare our results with a previous study of Supelec students that used a Support Vector Machine (SVM)algorithm [4] for the same aim.  READ THE COMPLETE REPORT

---

(click to read full report)