Comparison of machine learning algorithms for credit card fraud transaction prediction

Abstract

Nowadays, credit card transaction is widely used in commercial activities, the number of fraud transaction is increasing. Especially when e-commerce and online shopping have arisen explosively and COVID-19 pandemic spread, the fraud transaction increasing surprisingly. Detecting the credit card fraud transaction is of great importance for both banks and depositors. To verify the effectiveness of machine learning algorithms for this task, several models are leveraged and examined in this work. These models include naive Gaussian bayes model, logistic regression model, random forest classifier, convolutional neural network (CNN) and support vector machine (SVM). Besides, the effectiveness of pre-processing techniques, including undersampling and oversampling, is also conducted on the training dataset for validating their effectiveness on predicting whether a transaction is a fraud or not and their joint effort with different models. In this work, these models are implemented with European dataset of credit card fraud and the best method is selected for this application.