Fraudulent event detection in credit card data operations using SVM-RBF kernel function machine learning classification technique

Credit card operations are essential to human life as technological development has been overgrown in the last few years. Customers get connected with online operation systems easily due to the ease of the process of online systems. As technology grows rapidly the fraud associated with credit operations also grows rapidly along with technological developments. To find or identify such kinds of fraud actions, machine learning, and its classifier techniques play a very important part in identifying the patterns of fraud and legitimate actions of customers based on earlier operations. To identify the behaviour of operations it is necessary to study all previous operations executed by credit card holders to analyze the behaviour that they have executed, Further, they are classified as either fraudulent or legitimate operations. This paper proposed the radial basis kernel function integrated with the support vector machine's Kernel function model. The proposed methodology of radial basis kernel function for Support vector machine is compared and analyzed with like linear, sigmoid kernel functions to perform the fraudulent detection analysis. The execution analysis of this model is analyzed based on different performance metrics like accuracy, sensitivity also known as recall, specificity, etc. The model shows execution results of RBF kernel functions giving significant accuracy, recall, and specificity than that of other kernel SVM functions and existing machine learning classifiers. Therefore, the proposed model radial basis SVM is designed based on the features that use the Gaussian function instead of the sigmoid kernel functions as a basis function. The proposed RBF-based SVM function performs better than the other kernel methods and existing classifiers. The accuracy performance metrics of the proposed RBF -SVM are achieved at 95%, whereas sensitivity also known as recall is achieved at 89%, and the specificity performance metric is achieved at 90% which is better than other.