An Improved Support Vector Data Description Algorithm

In the real world, everything has its features, these feature is more or less, or important orunimportant. People can be determined by the features of things Category But when thefeatures of things, if people rely on traditional methods to classify things it is time-consumingand labor-intensive, and the classification accuracy is not high. Classification as a predictivemodel, classification accuracy is low or cost time is long, this forecast will become worthless.Have been proposed various classification model to predict things, SVM and SVDD hascertain advantages, and depending on the requirements, the improved application of these twoalgorithms in high-dimensional data to predict in many areas of the real life.Firstly, this paper studies the background and theorey SVM classification algorithm indata mining, analysis and summarizes the research status of the SVM various improvementsmethod. Secondly,On the basis of analysis and research support vector data description ofseveral improved algorithm, SVDD classifier is constructed essentially solving a quadraticprogramming optimization problem,and factors that affect the decision boundary of theSVDD algorithm is training samples of support vector.In order to improve the training speedof the SVDD algorithm,introduce of the K-means clustering and sample similarity,proposed abased Reduction Set and Tow training Support Vector Data Description algorithm. Thealgorithm uses the k-means clustering and sample similarity interval divided training sampleset into multiple subset,and randomly selected training samples in the subset as a trainingsubset to training a SVDD classifier decision boundary,then use the classifier decisionboundaries to identify the original training set possible support vector constitute a newtraining subset, to train another SVDD classifier decision boundaries.Finally, on threedifferent dimensional breast-cancer data sets, respectively SVDD algorithm, RSVDDalgorithm RSTSVDD algorithm carried out experiments, and compared their performance.Experimental results show that on the basis of essentially the same classification accuracy,RSTSVDD algorithm has higher training speed.