An Algorithm of Sphere-Structure Support Vector Machine Multi-classification Recognition on the Basis of Weighted Relative Distances

摘要

Theories on sphere-structure support vector machine (SVM) and multi-classification recognition algorithms were studied in the first place, and on this basis, in view of the issue of the difference in the hypersphere radiuses resulted from the difference in the quantity of the training samples and the discrepancy in their distributions, the concepts of relative distance and weight were introduced, and subsequently a new algorithm of sphere-structure SVM multi-classification recognition was proposed on the basis of weighted relative distances. Accordingly, the data from the UCI database were used to conduct simulation experiments, and the results verified the validity of the algorithm propose.