A feature extraction method for fraud detection in mobile communication networks

To improve the fraud detection accuracy by SVM (support vector machine), a feature extraction method named GPCA based on IG (information gain) and PCA (principal component analysis) is proposed. It analyzes the data on CDR (call detail record), customer information, paying and arrear information etc. in mobile communication networks, and then the data can be used by the SVM classifier to build the fraud detection model and the user can predict the potential fraud customers. Despite of its simplicity, GPCA outperforms some of the most popular feature extraction methods such as BS (bivariate statistics), IG and PCA in predicting accuracy and training time. To get the higher predicting accuracy, a binary SVM using RBF (radial basis function) kernel is used. The experiments show that the classifier with GPCA has fine predicting accuracy.