Enhanced SMOTE & Fast Random Forest Techniques for Credit Card Fraud Detection

Abstract

The credit card payment industry has steadily grown over the last decade with ascend and extensive use of the internet. Credit card is a means of payment without getting cash in hand for purchasing or selling products or services. It is an easy way of automatically providing credit to a customer. Fraud in the credit card industry has turned out to be a big issue. Credit card fraud results in damages of billions of dollars for online traders, credit card users & it is a serious concern of great significance.Identification of financial crime and risk management are important for banks and business organisations. The predictive models for the detection of credit card fraud are in active use in practise, there are relatively few documented studies on the use of data mining approaches for the detection of credit card fraud, likely due to lack of study data available and reluctance to share it for a potential intangible loss.

The proposed approach suggests a smart credit card fraud detection model which can detect fraud from extremely imbalanced and unspecified credit card transaction dataset. The proposed fraud detection model is based on 3 phases, including pre-processing in which redundant attributes are eliminated, and also ranks characteristics using the Fast Random Forest model by their significance. After pre-processing, the issue of class imbalance is addressed by the Enhanced SMOTE (Synthetic Minority Oversampling Technique) technique, and the final step is to classify whether Fast Random Forest performs a new transaction is either fraudulent or legitimate. In this research work, we have used UCSD FICO (University of California San Diego Fair Isaac Corporation) data mining competition dataset for performance assessment of presented approach.