Credit Card Fraud Detection Using Feature Fusion-based Machine Learning Model

Abstract

Credit card fraud is a very common means of financial fraud at present. In order to reduce social and personal economic property losses, this study aims to propose a model to predict credit card fraud, so as to detect possible fraudulent transactions in the transaction process. In this paper, the data set to simulate fraudulent transactions be used, and construct the model after processing the characteristic data and selecting the features. In the model construction part, four algorithms are used as the trained classifiers, which are K Nearest Neighbor algorithm, Bgging algorithm, Logistic Regression algorithm and Gaussian Bayesian algorithm. First, the four algorithms are used to train the data set respectively, and then the parameters of each classifier are adjusted respectively, so that each individual classifier achieves the optimal performance. On this basis, a new model is constructed by using the method of model fusion. The classifier constructed by K Nearest Neighbor algorithm, Bagging algorithm and Gaussian Bayesian algorithm is used as the primary learner, and the classifier constructed by logistic regression algorithm is used as the secondary classifier to build a fused model as a whole. The final experimental results will give the classification performance of the five models, including individual models. The results show that the performance of the model after integrating the four individual models is better than that of the single model, which can provide accurate feedback for credit card holders on whether there is fraud.