Comparing the Performance of Multivariate Hotelling’s T2 Control Chart and Naive Bayes Classifier for Credit Card Fraud Detection

Abstract

Credit card is a transaction tool using a card which is a substitute for legitimate cash in transactions. The use of computer technology is needed for various kinds of electronic transactions. In the world of technology, the term machine learning is not new and technological developments are increasingly rapid in recent years. Statistical process control method (SPC) is one of the measuring instruments used to improve the performance of public services. Hotelling T^2 control chart is a method in SPC that can be used to control the process. Methods that are widely used in the detection and classification of documents one of them is Naive Bayes Classifier (NBC) which has several advantages, among others, simple, fast and high accuracy. Those two methods will be used to detecting o2utlier of this dataset. The study used the credit card fraud registry with some PCA as independent variables. The size of fraud transaction is very small which represented only 0.172% of the 284,807 transactions. This research will use Area Under Curve (AUC) as the performance goodness test. A comparison of the accuracy of NBC and Hotelling's T2 predictions shows that the performance of the T2 Hotelling method is better in detecting outliers than the NBC method