摘要
提出了一种应用遗传算法优化子空间的SVM分类算法GS-SVM。该算法首先改进样本选择策略,采用基于置信度和凸包的样本选择方法,考虑类间距离和样本分布等因素,选择典型代表样本作为SVM的新训练集;然后采用矩阵式混合编码方式,利用遗传算法一并优化代表样本的特征子空间和SVM分类参数,并根据特征优化后的代表样本,构建SVM分类模型。在UCI的11个数据集上进行的仿真实验结果表明,该算法在大部分数据集上均可获得较小的样本规模和特征维数,以及较高的分类精度。
This paper presented a new GA based Subspace classification algorithm for SVM(GS-SVM). A modified sample selection method is adopted to select a subset of training data based on both the confidence and the convex hull. Then the representative samples are selected to train the SVM models by considering the distances between classes and the sample distribution. The algorithm adopts the matrix-form mixed encoding. Genetic algorithm is used to optimize the feature subspace of representative samples and the classification parameters of SVM simultaneously. The SVM classifi- cation model is produced based on the representative samples with the optimized feature subspace. Experimental results on eleven UCI datasets illustrate that the proposed algorithm is able to select both smaller sample subset and feature size,and achieve higher classification accuracy than the traditional classification algorithms.
出处
《计算机科学》
CSCD
北大核心
2013年第11期255-260,275,共7页
Computer Science
基金
国家自然科学基金项目(61202030)资助
关键词
子空间分类
遗传算法
支持向量机
样本选择
凸包
Subspace classification,Genetic algorithm, Support vector machine, Sample selection, Convex hull