Support Vector Machine for Frequently Executed Method Prediction

Frequently executed methods have major impact in increasing the performance of CPU as well as in code optimization to address the best set of optimization. In this study it is proposed to develop a machine learning based predictive model using Support Vector Machine (SVM), classifier for identification and prediction of frequently executed methods in a program. Since kernel functions play an important role in performance of the SVM and choosing suitable kernel function is still an area of research, here the performance of various kernels of SVM classifier is evaluated and compared. It is found that when trained with ten static program features, the derived SVM model using Radial Basis Function (RBF) kernel function predicts the best accuracy of 65% using SPEC CPU 2000 INT dataset. When the same data set was compared with Linear kernel, Polynomial and Sigmoid kernel the prediction accuracy is found to be 40%, 45% and 15% respectively.