Fraudulent Detection in Credit Card Transactions using Radial Basis Function Kernel Method based Support Vector Machine

Abstract

Usage of online transactions in everyday life of human being has been increasing since last few years due to technological advancements. Due to easy and convenient online payment transaction systems, users are getting connected so quickly. Credit card fraud is also rapidly growing along with the technological development. To identify or detect such fraudulent act, machine learning plays important role to check the patterns of a normal and fraudulent behavior of user over the past transactions. To detect the transaction status it is important to analyze all past transaction record done by credit card customer to indentify the pattern he or she used, we can further classify as either legitimate or fraud transactions In this paper, support vector machine's Kernel functions model is proposed. Four kernel functions like linear, sigmoid, polynomial and RBF kernel function is used for fraud detection analysis. The classifiers like KNN, Naïve Bayes, Logistic Regression, decision Trees as existing classification models. The non linear performance is measured with SVM Kernel functions (polynomial, sigmoid, RBF). The  performance evaluation of these techniques based on accuracy, sensitivity, specificity etc. The performance result shows all kernel gives better accuracy, Sensitivity and specificity than existing classifiers. Hence the new proposed model sigmoid-RBF is developed based on the RBF uses composite of sigmoid functions in place of the Gaussian function as the basis function. The proposed sigmoid- RBF function gives significant performance over the other kernel functions. The accuracy level is 96%, sensitivity is 90% and specificity is 92% which is effectively better than other SVM kernel functions.