The Research On The Extension And Application Of SVM Classification

Support Vector Machine (SVM) is new in Data Mining, which's usually used as a classifier with intact theory background, strong adaptability, and excellent generalization function. Also the optimization is for overall situation. In some challenged applications at present, it has gained the best performances till now. Aimed to resolve practical questions and to expand the applicability of SVM, the primary content of this paper is to improve the method of SVM classification algorithmic and to expand its application, on the basis of studying classical SVM algorithmic.Started from the background theory of SVM, this paper analysed the research results of SVM training algorithmic until now and its applications in Data mining field (especially in classification algorithm). Also this paper analysed the achieved direction of SVM in Data Mining field and revealed SVM classification strategy. Aimed at loosening the restricting about the applications of SVM, this paper brought forward two methods to improve SVM classification performances. It has been confirmed that these two methods had improved the classification performances in the application of spam-email filtration in some degree:1. Making use of the unmarked data to improve classification performance. It brought forward a method in the situation of insufficient training samples. It chose sample data to mark class labels voluntarily, and then added to the SVM training set. By setting threshold value, it made sure that the chosen samples have high validity.2. Improving classification performance based on the thought of integrated learning.It brought forward a method called KSU based on integrated learning. The optimization classification surface could not give a very good category decision to some samples, so we used K-Nearest-Neighbor method as an aid and used the support vectors ascertained by SVM as K-Nearest-Neighbor training set. This method can improve the classification accurate rate.Using this classification model for spam-email filtration, this paper analysed why spam emails existed, summarized the usual methods for spam-email filtration and expounded detailedly the procedure of spam-email filtration methods based on web content. Compared with classical SVM that maked decision only by the optimization classification surface, the experiment indicated that this method had shown enough superiority in the samples scale suitable for use. And it was very obvious that it had been better in improving accurate rate and reducing the sensibility to Kernel and to the parameters of SVM. This model increased the time complicacy to a little degree, but it could assure the same space complicacy as classical SVM. Accordingly, it's efficient and practical.