An improved SVM classification method based on GA

Support. Vector Machine (SVM) is a new kind of machine learning method, which constructs an optimal hyperplane from a small subset of samples, near the boundary, named as support vectors. But in large scale problems choosing these, support vectors by SVM involves a large scale quadratic programming and is thus of high computational cost, which is regarded as one of the bottle-neck for its application. In this paper we present a new algorithm developed from classical SVM, which use Genetic Algorithm (GA) to preprocess the initial training data for a small subset, which may include most of the support vectors, then train a classical SVM with this subset to obtain a classifier. The new method combines the benefits of the generalization ability, of SVM and the capability of GA in dealing large-scale data set. Our experimental results suggest that GA-SVM may have a better performance than SVM in both training speed and generalization accuracy.