关键词: 支持向量机, 超球体, 不等距, 最大间隔

Abstract: For different kinds of samples，the following cases are often confronted：The scopes of sample-distribution differ largely，the degrees of harm for misjudgment of sample class are different，or，the numbers of different kinds of samples are various，pointed on which，SVM（NMS-SVM） based on the non-equidistant margin hypersphere is proposed.Aim at optimizing the maximal margin，SVM introduces distance ratio parameter and adjusts the distance between the optimized classing surface and the two classes.Through the simulation experiment of classification of sample collection by UCI corpus，the validity of SVM is proved by comparing the precision of classification with the normal hypersphere arithmetic as well as the maximal margin hypersphere arithmetic.

Key words: Support Vector Machine（SVM）, hypersphere, non-equidistant margin, maximal margin