Research Of Support Vector Machine Multiclassification Model Based On Granular Computing

Statistical learning theory, as a machinery learning strategy in a set of small samples, has provided a unified structure to solve the problem of learning with a limited set of samples. Support Vector Machine(SVM), a universe leaning method based on the statistical learning theory, is with many advantages such as global optimization, strong adaptation, full theory and generalization. SVM improves the general performance according to the structural risk minimization, which is more complete in dealing with the two-classification, non-linearity, small set of samples and has solved the curse of dimensionality and over-learning.With the development of science and technology, the increasing disturbing elements as variety, uneven distribution and large amount of information have made the multiclassification more difficult. The thesis researches on a new type of Multi-classification model, which combines Granular Computing(GrC), Huffman tree, SVM to improve the efficiency and precision of the classification.First, the GrC theory was added to the multi-classification. The specific problems were defined and conversed by the grain of ternary theory. The granular structure was constructed with the hierarchical thinking in the GrC. The granular size was coarsened and refined. The statistics was processed and analyzed from different angles so as to deal with problems of the large amount of statistics and low computing speed.Second, Huffman tree based on the granular ranking was constructed to deal with the unevenness of within-class scatter and the low efficiency of classification. For the weighed path length is the shortest in the Huffman tree, the classification can be ascribed with the minimum time. The construction of Huffman tree was achieved with the VC++6.0. Then the Huffman tree codes were gained.Finally, the grain in the decision tree was analyzed to design different SVM classifier corresponding to every grain. The Global classification model was established. The model training and simulation were realized with MATLAB. It ensures the model being scientific and effective testifying with examples. Figure 17; Table 9; Reference 61...