A Note on Parameter Selection for Support Vector Machines

Parameter selection greatly impacts the classification accuracy of Support Vector Machines (SVM). However, this step is often overlooked in experimental comparisons, for it is time consuming and requires familiarity with the inner workings of SVM. Focusing on Gaussian RBF kernels, we propose a grid-search procedure for SVM parameter selection which is economic in its running time and does not require user intervention. Based on probabilistic assumptions of standardized data, this procedure works by filtering out parameter values that are not likely to yield reasonable classification accuracy. We instantiate this procedure in the popular WEKA data mining toolbox and show its performance on real datasets.