Research And Design Of Multi-classification Model Based On Support Vector Machine

Optimizing for support vector machine(SVM) classification model is an important research hot spot in the area of pattern recognition because the SVM’s recognition ability is much better than the traditional machine learning method in dealing with a small set of high-dimensional sample data. But SVM must extend the binary classification to the multi classification while it is used in multi classification, which often results in the problem of decisions blind spot, the data set tilt, and the Consistent-diverse issues. Finally, it would lead to the decision deviation and even wrong decision. Thus, in order to solve the above problems, this project will construct a multi-class classification model from a new view, which will discard the existed method1vs1,1vsA, and ECOC.The main contents of this thesis are to design a simple and effective method to obtain the linearly separable situation of the sample set and propose a multi-classification model building method based on the previous method in order to optimize recent multi-classification effectiveness and efficiency and make up for the lack of the algorithms that include1vs1.1vsA, ECOC and MOC. The BreastTissue dataset is taken as an example to describe the algorithm that has been implemented. In additional, we use some UCI datasets, for example. Iris, BreastTissue, Statlog (Image Segment) to verify the effectiveness and efficiency of the multi-classification model based on SVM compared with the aboved multi-classification algorithms. Finally, we conclude the recent research work and point out the next step of the related research.