核主成分马田系统及其应用

Kernel principal component Mahalanobis-Taguchi system and its application

为解决传统马田系统无法有效识别非线性数据问题,提出构建核主成分马田系统.该方法通过核马氏距离的构建,将马田系统和核主成分分析两种方法融合在一起,使其不但具备非线性数据识别能力,还具备数据降噪能力.实例验证表明:核主成分马田系统在降维率、识别准确率、不平衡数据处理能力、特异度和灵敏度等方面不仅优于传统马田系统,还优于BP神经网络、随机森林和逻辑回归等方法.同支持向量机递归特征消除法相比,在识别准确率和不平衡数据处理能力方面,两种方法性能接近,但当数据复杂度较高时,核主成分马田系统要优于支持向量机递归特征消除法,并且核主成分马田系统需要的维数较少,也不需要反复迭代.

In order to solve the problem that the nonlinear data cannot be identified effectively by the traditional Mahalanobis-Taguchi system(MTS),the kernel principal component Mahalanobis-Taguchi system(KPCMTS) is proposed.The KPCMTS integrates MTS and kernel principal component analysis through the construction of kernel Mahalanobis distance.So the KPCMTS can not only identify the non-linear data,but also reduce the data noise.Experimental results demonstrate that the KPCMTS is not only superior to the traditional MTS on dimension deduction rate,accuracy,g-means,specificity,and sensitivity,but also superior to BP neural network,random forest,and logistic regression.Compared with support vector machine recursive feature elimination(SVM-RFE),the performance of both methods is similar on accuracy and g-means.However,when the data complexity is high,the KPCMTS is superior to the SVM-RFE,and the KPCMTS requires fewer dimensions and does not need repeated iteration.