A methodology using EMO for parameter estimation of SVM kernel function

The problem of kernel parameterization for support vector machines (SVMs) is considered. This paper tried to apply EMO to the parameters estimation problem of SVMs, and we investigated which combination of objectives is proper for this problem. We examined the performance of SVMs classifier with using not only cross-validation(CV) case, but also inverted CV which the ratios of training data and external test set is swapped. Inverted CV was used for the performance estimation in which a limited amount of sample can be used for training data. In our experiment, we used the two different kinds of problems; graphic two dimensional problems and benchmark data sets taken from the UCI Machine Learning Repository. Through experiments, we investigated the synergistic and effectiveness of an objective combination.