A ROC Curve Method for Performance Evaluation of Support Vector Machine with Optimization Strategy

Support vector machine is the highlight in machine learning. Also, performance evaluation and parameters selection for SVM model become an important issue to make it practically useful. In this paper, after investigating current evaluation index for pattern recognition, we introduced Receiver Operating Characteristic Curve into the performance evaluation. Area under Receiver Operating Characteristic Curve is applied to the model evaluation, model performance of SVM and RBFN is compared. Also optimal operating point of ROC is adopted to the optimization of SVM within the kernel parameters and penalty factor, and the optimization is performed by seeking of optimal operating point. Pattern recognition experiment with UCI dataset shows that ROC method is an effective approach for performance evaluation and optimization of SVM.