Penalized Matrix T Mixed Model and Its Application in Provincial Economic Classification
In order to fully consider the characteristics of the matrix data and the correlation within the data,this paper con-structs a matrix T mixed model based on the matrix T distribution and its penalized model to study clustering.A penalized likeli-hood estimation algorithm is proposed by applying an adaptive nuclear parametric low-rank penalty to the mean matrix compo-nents on the likelihood function of the matrix T mixed model and applying the ECM algorithm.At the same time,an improved BIC model selection criterion is also proposed to select the optimal number of mixed models and tuning parameters,which in turn auto-matically implements low-rank estimation to achieve accurate clustering by adaptive nuclear parametric thresholding.Finally,the numerical simulation study is compared with existing methods to verify the usefulness of the method;the developed penalized mixed model is applied to the study of the classification of economic development levels in Chinese provinces,and more accurate clustering results are obtained.
matrix T distributionmixed modeladaptive nuclear normECM algorithmsingular value threshold
万方数据
维普
