一种新的基于NGA/PCA和SVM的特征提取方法

Novel Feature Selection Method Based on NGA/PCA and SVM

提出了一种使用小生境遗传算法(NGA)和主成分分析(PCA)对支持向量机(SVM)进行封装的方法来选择特征子集。该方法首先使用PCA得到特征向量,然后产生若干随机特征向量子集,从而得到新的特征空间,将所有训练样本映射到这个特征空间来训练支持向量机,再使用支持向量机的半径间隔方法对每个特征向量子集的性能进行评价,最后使用小生境遗传算法来共享适应度,以及进行选择、交叉和变异操作得到新的特征向量子集,重复这个过程直至得到最优的特征向量子集。使用UCI数据集进行了相关的实验,实验结果表明了该方法可以减少特征的数量以及提高分类正确率。

The performance of support vector machine （SVM） highly depends on feature subset, and there are some shortcomings of the classical filter approach to feature selection, The objective of this research was to optimize the feature subset, For there are some shortcomings of simple genetic algorithm, so niche genetic algorithm was introduced, The eigenvector were computed by principle component analysis （PCA）, and eigenvector selection was performed by generating random collections of eigenvector, and the feature was obtained using these eigenvectors, and support vector machine was used as an evaluation of the eigenvector subset, and did the niche genetic algorithm, and some new eigenvector subset generated, and loop the above step till the stop condition satisfied. Experimentation was carried out using UCI data, and the experimental results show that the new algorithm significantly improves the classification accuracy,