In order to solve the problem that traditional Mahalanobis-Taguchi system cannot process interval data,an interval Mahalanobis-Taguchi system(IMTS)is proposed.The key of the IMTS is the construction of scaled interval Mahalanobis distance.Firstly,the interval sample vector is decomposed into upper bound sample vector and lower bound sample vector,and then the upper bound Mahalanobis distance and lower bound Mahalanobis distance are calculated respectively.Finally,the mean value is defined as the interval Mahalanobis distance,and the properties of the interval Mahalanobis distance are proved,and the scaled interval Mahalanobis distance is defined according to the property of the interval Mahalanobis distance.Six interval data sets are generated by simulation to verify the performance of the IMTS.The verification results show that the IMTS has strong linear recognition ability,dimension reduction ability,unbalanced data processing ability and certain nonlinear recognition ability,and its performance is less sensitive to the changes of interval length.Finally,the detailed calculation steps of the IMTS are given,and compared with seven pattern recognition methods for poverty recognition,including logistic regression,decision tree,support vector machine,random forest,naive Bayes,K-nearest neighbor and neural network,respectively.The comparison results show that the overall recogni-tion performance of the IMTS is better than the above methods.In particular,the poverty-returning recognition accuracy of the IMTS is higher,which verifies the superiority of the IMTS in outlier recognition.
NETL
NSTL
万方数据
