基于微粒群算法和支持向量机的特征子集选择方法

在模式分类系统中,大量无关或冗余的特征往往会降低分类器的性能,因此需要特征选择.本文提出了基于离散微粒群(BPSO)和支持向量机(SVM)封装模式的特征子集选择方法,首先随机产生若干种群(特征子集),然后用BPSO算法对特征进行优化,并用SVM的10阶交叉验证结果指导算法的搜索,最后选出最佳适应度的子集对SVM进行训练.两个UCI机器数据集(户外图像和电离层)的实验结果表明了提出算法的有效性.

BPSO-SVM Wrapper for Feature Subset Selection

In pattern classification system,many irrelevant and redundant features will lessen the performance of classifiers.So it is important to select features.This paper proposed a discrete binary version of particle swarm optimization-support vector machines (BPSO-SVMs) wrapper mode feature selection algorithm.At first,a population of particles (feature subsets) was randomly generated.Then BPSO algorithms searched the feature space guided by the result of SMVs' 10-fold crossover validation.After numbers of iteration,the best fitness feature subset was selected out to train the predictor.Experiments on two datasets (Segmentation and Ionosphere) in UCI machine learning repository confirm the effectiveness of the proposed strategy.