摘要
提出了一种基于改进遗传算法的特征选择算法。该算法以支持向量机分类器的识别率作为特征选择的可分性判据,对传统遗传算法的交叉和选择操作进行了改进,实现了指定数目的特征选择。而且算法在特征选择的过程中,还同时优化了支持向量机分类器的两个参数。实验数据的特征选择实验表明,提出的算法仅以损失2.7%识别率的代价,得到的特征维数却是传统遗传算法的1/5,极大地简化了分类器设计的复杂度。
A novel Support Vector Machine（SVM） feature selection algorithm is proposed based on modified genetic algorithm.The recognition rate of SVM is used as separability criterion,and the crossover and mutation operation of traditional Genetic Algorithm （GA） are modified in the proposed algorithm.Two parameters of SVM are optimized during feature selecting.The results of experiment show that,the novel algorithm losses the recognition rates of 2.7 percent,but the length of feature vector is one fifth of traditional ones,and simplifies the complexity of classifier.
出处
《计算机工程与应用》
CSCD
北大核心
2009年第29期28-30,共3页
Computer Engineering and Applications
基金
国家重点实验基金(No.51444030105JB1101)
关键词
特征选择
支持向量机
遗传算法
feature selection
support vector machine
genetic algorithm
