Credit Card Fraud Detection

Abstract

Worldwide billions of dollars per year goes into vain because of credit card fraud which is a major on growing problem. High tech advanced classification methods provide the ability to detect these fraudulent transactions without much disturbance to legal transactions. However, some key risks are involved namely, (a) imbalance learning and (b) concept drift. This paper involves some hybrid sampling for handling class imbalance at the data level and algorithm level which is being examined on European credit card transactions over a period of two days. The results are compared with the three main algorithms which have high performance in the task of fraud detection: (a) linear support vector machine, random forest, and K-NN. Thus, it has been proven from the results that sophisticated generative sampling methods will fall short in generalizing the minority class in the presence of extreme class imbalance. The work is implemented in Python. The execution of the method is evaluated on the basis of accuracy. The outcomes show of optimal accuracy for support vector machine, k-nearest neighbor and random forest classifiers are 5, 79, and 84% separately. The similar outcomes demonstrate that random forest performs superior to support vector machine and k-nearest neighbor algorithms.