Study And Application On The Searching Of Support Vectors Based On SVDD

Support vector machine(SVM),which is based on statistical learning theory, is one ofmachine learning algorithms was presented in1990’s by Vapnik.SVM make the actual riskminimized through the rule of structural risk minimization, and it can get optimal result even ifthe samples is limited.SVM has high generalization ability, and over come the drawbacks thatsome traditional machine learning algorithms get the local optimal solution easily. All of theseadvantages make SVM became a very good machine learning algorithm.Support vector data description was presented in the last century by Tax, it focuses on theshape of dataset. A good description should cover all target objects but includes no superfluousspace. The boundary of a dataset can be used to detect novel data or outliers. SVDD is inspiredby the SVC.A lot of algorithms which are based on SVDD were presented recently.The proposed method is based on SVDD, it is applied to multiclass problems. We mainlyfocus on the training speed, and the decision function which is used for describing the boundaryof dataset. The proposed method also uses the domain density description, we only need toconstruct model of support vector learning for one class. It is similar to Bayesian decision rule.The results of experiments show that the proposed method can train faster and reduce the size ofdata for the description, and faster than the traditional methods in multiclass.