基于混合特征选择的轻度认知功能障碍的诊断分类

摘要

为了有效提高轻度认知功能障碍(Mild Cognitive Impairment,MCI)的诊断分类效果,提出了一种基于Relief算法和支持向量机回归特征消除SVMRFE算法的混合特征选择方法Relief-SVMREF,该算法首先利用Relief算法去除无效特征,同时针对Relief算法无法去除冗余特征的问题,本文利用皮尔逊相关系数对选择出的特征进行冗余分析,去除冗余特征.最后利用SVMRFE算法对选出的特征进行排序,得到最终排序系数.对得到的特征排序采用留一交叉验证方法获取最优子集,再用SVM分类识别.实验结果表明该方法能够取得更好的分类效果.%In order to improve the diagnostic effect of mild cognitive impairment(Mild Cognitive Impairment,MCI), this paper presents a mixture feature selection method Relief-SVMREF algorithm based on Relief and SVMRFE algorithm. Firstly we use the Relief algorithm to remove invalid characters. The relief algorithm does not remove the redundant features, we use the Pearson correlation coefficient to remove the redundant features. Finally, the feature is sorted by SVMRFE algorithm, and the final ranking sequence is obtained. We obtain the optimal subset with the left one cross validation method, and then we use the SVM to do classification. The results show that the method can get better results compared with the two method along.