【摘要】 在本文中,以阿尔茨海默症患者的脑皮层厚度作为数据集,利用mRMR特征选择方法对SVM-RFE特征选择方法进行改进,以提高轻度认知障碍人群和正常人群分类的准确率.SVM-RFE特征选择方法是根据SVM在训练时生成的权向量来构造排序系数,并在每次迭代时去掉排序系数最小的特征.该方法只考虑到特征与类标的相关性未能考虑到特征间的冗余性,鉴于此,在生成权向量后,引入mRMR里计算相关的算法来重新构造排序系数,并在每次迭代时去掉排序系数小的特征.实验使用留一交叉验证进行评估,结果表明本文方法要优于SVM-RFE特征选择方法、mRMR特征选择方法和F-score特征选择方法.更多还原

【Abstract】 In this paper,the cortical thickness of Alzheimer’s disease was taken as the data set,combined with mRMR feature selection method toimprove the SVM-RFE feature selection method,therebyimproving the accuracy of mild cognitive impairment and normal classification. SVM-RFE feature selection methodconstructs the sorting coefficientsaccording to the weight vector generated by SVMduring training and removes the feature with the smallest ordering factor at each iteration. This method only takes into account the correlation between features and class labels and fails to take into account the redundancy among features,in viewof this,after generating the weight vector,introductinga calculation inthe mRMR algorithm to reconstruct the sorting coefficientsand remove the feature with the smallest ordering factor at each iteration. The experimentwas evaluatedby leave-one-out cross validation,and the results showthat this method is superior to SVM-RFE feature selection method,mRMR feature selection method and F-score feature selection method.更多还原