Credit Card Fraud Detection using Data Pre-processing on Imbalanced Data - both Oversampling and Undersampling

Abstract

Online or Offline purchasing using credit or debit card has been popularized rapidly for its ease and flexibility of use. With the growing rate of online purchasing, fraudulent transactions have also increased with an alarming rate. Researchers have investigated the reason and remedies of these kind of acts. However, credit card fraud detection can be categorized into two sections in application fraud and behavioral fraud aspects. While an application is being compromised by the cyber-crimes, then, the user behavior analytic system has come to the limelight in recent days. In this research, a fraud detection system has been built, which can distinguish three types of user from pre-processed transaction logs. To prepare a balanced dataset, undersampling of majority class and synthetic minority over-sampling technique on minority class applied on feature vectors. K-nearest neighbor is used to distinguish the legal, fraudulent and enigmatic users. This system outperforms current state-of-art techniques to get behavioral analytics.