Design of a Model in Machine Learning For Credit Card Fraud Detection

Abstract

In e-commerce, fraud has become a major problem, and much effort is being invested into identifying and preventing it. Currently, fraud detection and prevention systems are only able to detect and prevent a small percentage of fraudulent transactions, resulting in billions of dollars in losses. Because online transactions are likely to grow dramatically in the coming year, better fraud detection and prevention is critical. We provide a datadriven strategy for estimating the likelihood of a fraudulent or legal transaction based on machine learning techniques applied to large data sets. To predict the likelihood of a customer's next transaction being fraudulent, the algorithm was trained using past e-commerce credit card transaction data. Random Forest, Support Vector Machine, Gradient Boost, and combinations of these are used to compare the performance of supervised machine learning approaches. The problem of class imbalance is considered before the model is trained on a classifier, and methods such as oversampling and data pre-processing are employed.