摘要： 针对采用支持向量机进行分类的特征子集选择问题，提出一种改进的基于梯度向量的特征评测算法。该算法在核特征空间中，利用数据点到分类超平面的距离函数的梯度向量对各个特征的重要性进行排序，省去了已有算法中计算梯度向量与各个坐标轴夹角的过程，实验结果表明，该算法简化了已有的基于角度的特征选择方法，并且结果保持一致。

关键词: 支持向量机, 特征选择, 核函数

Abstract: An improved feature selection algorithm based on gradient is proposed, aiming at the problem of feature subset selection when applying Support Vector Machine(SVM) to classification. The algorithm ranks the features by the criterion of the local gradients of distance function, measuring the distance from a datapoint to the classification hyperplane in the kernel-induced feature space, and eliminates the calculation of the angles of gradients with every axis. Experimental results show that the proposed algorithm simplifies the method of feature selection based on angle calculation and keeps the results consistent.

Key words: Support Vector Machine(SVM), feature selection, kernel function

中图分类号: 

• TP311.13