基于核主成分分析的特征提取方法

Feature extraction based on Kernel Principal Component Analysis

为了证实核主成分分析在特征提取中的优越性,利用支持向量机作为分类器,以主成分分析和核主成分分析作为特征提取的工具,以分类器的分类性能作为方案优劣的评判标准设计了六种实验方案进行实验分析。实验数据表明,对特征选择后的数据集利用主成分分析和核主成分分析进行特征提取,可将数据投影到一个更低维的特征空间,实现数据维数的约简和分类器性能的提高。同时还发现,在对数据进行特征提取的能力上,核主成分分析优于主成分分析。

In order to confirm the advantage of Kernel Principal Component Analysis （KPCA） in feature extraction,six schemes were designed to undertake experiment analysis, which used Suport Vector Machine （SVM） as a classifier, KPCA and Principal Component Analysis （PCA） as a tool for feature extracxtion and the classifier＇s performance as a criterion to evaluate each experiment scheme. The experiment result indicated the dataset had been carried out feature selection could be mapped to a lower feature space by using KPCA and PCA for feature extraction. As a result, the dimension of dataset was reduced and the performance of the classifier was improved. At the same time, we could find KPCA was more effective than PCA in feature extraction.