CAJ | 학술논문

　　传统模糊支持向量机在原始空间计算模糊权重，复杂度较高，而且不利于模糊权重的精确赋值。针对该不足，本文提出了一种基于中心向量投影的模糊权重赋值方法。该方法首先计算两类原始样本的样本中心，然后利用两个样本中心所在直线作为投影直线，计算原始样本投影，最后根据投影分布确定相应原始样本的模糊权重。该方法通过降维有利于噪声的判断，使得模糊权重赋值更加准确，而且计算复杂度较低。在文本分类数据集上的成功应用说明本文算法可以有效地提高模糊支持向量机的分类性能。
　　전통모호지지향량궤재원시공간계산모호권중，복잡도교고，이차불리우모호권중적정학부치。침대해불족，본문제출료일충기우중심향량투영적모호권중부치방법。해방법수선계산량류원시양본적양본중심，연후이용량개양본중심소재직선작위투영직선，계산원시양본투영，최후근거투영분포학정상응원시양본적모호권중。해방법통과강유유리우조성적판단，사득모호권중부치경가준학，이차계산복잡도교저。재문본분류수거집상적성공응용설명본문산법가이유효지제고모호지지향량궤적분류성능。
Traditional fuzzy support vector machines compute the fuzzy weights in the original space, which is with a high computational complexity. Moreover the fuzzy weights can not be compute accurately. For the shortage of traditional fuzzy support vector machines, this paper proposed a fuzzy SVM algorithm based on mean vector projection. Firstly, the means of the two classes are computed. Then the line though the two means is regarded as the projection line, and thus the pro-jection of all the training samples can be computed. Finally, determine the fuzzy weights of original samples according to the distribution of projections. Through dimensionality reduction, the fuzzy weights can be assignment more accurate and the process is with a low computational complexity. Experiments on text classification data set clearly reflect the effec-tiveness of the proposed algorithm.