A Survey on Fraudulent Transaction Detection using Random Forest

Abstract

In the evolution of the electronic money system, frequent transaction fraud has been a shadow behind the prosperity. It not only endangers the property security of users, but also hinders the development of digital finance in the world. With the development of data mining and machine learning, some mature technologies are gradually applied to the detection of transaction fraud. This paper proposes a transaction fraud detection model based on random forest. The experimental results of IEEE CIS fraud dataset show that the method of this model is better than the benchmark model, such as logistic regression, support vector machine. Finally, the accuracy of our model reached 97.4%, and the AUC ROC score was 92.7%. The random forest classifier is composed of a group of decision trees. Each tree is generated by independent sampling random vectors, and each tree votes to find the most popular category to classify the input. Random forest has both sample randomness and characteristic randomness, and its generalization performance is superior. At the same time, random forest has good processing ability for high-dimensional data sets, which is very suitable for IEEE CIS data sets. It can process a large number of inputs and determine the most important characteristics. Therefore, further feature mining is carried out on the data extracted by RFECV.