The Model Selection for Semi-Supervised Support Vector Machines

Abstract

Model selection for semi-supervised support vector machine is an important step in a high-performance learning machine. It is usually done by minimizing an estimate of generalization error based on the bounds of the leave-one-out such as radius-margin bound and on the performance measures such as generalized approximate cross-validation empirical error, etc. In order to get the parameter of SVM with RBF kernel, this paper presents a linear grid search method, which combines grid search and linear search. This method can reduce the resources required both in terms of processing time and of storage space. Experiments both on artificial and real word datasets show that the proposed linear grid search has the advantage of good performance compared to using linear search alone.