Research On Panel Data Clustering Based On Gaussian Mixture Model

Two main problems are concerned while facing with the clustering problem of panel data.One is similarity measurement of different samples,the other is the selection of specific clustering algorithm.Previous researches managed to measure such similarity using linear technology and succeed sample clustering using hard clustering algorithm.However,a nonlinear relationship always exists among variables due to the dynamic and structural characteristic of the panel data.Besides,the uncertainty of clustering can hardly be measured through hard clustering algorithm.Therefore,this paper is aimed to study the above problems.Four main contents are included in this paper.Firstly,finish the pre-treatment of nonlinear panel data using Kernel Principal Component Analysis(KPCA).Secondly,in order to estimate the uncertainty of samples clustering,this paper introduces a Gaussian Mixture Model(GMM)which is a kind of soft clustering algorithm.Thirdly,conduct an empirical analysis demonstrating the practicability of the proposed method.Fourthly,design a control experiment proving the effectiveness of the proposed method.The main conclusions of this paper are as follow.Firstly,the proposed panel data clustering method based on KPCA algorithm and GMM can be used as a systematic method of panel data clustering,which possesses certain practical application value.Secondly,KPCA algorithm works better than normal liner algorithm in extracting the nonlinear feature of panel data.Thirdly,the soft clustering algorithm based on GMM results a similar outcome as hard clustering algorithm,providing more descriptive information at the same time.The main conclusions of this paper can be concluded as follow.Firstly,put forward a feasible and effective panel data clustering method which extends the application range of cluster analysis.Secondly,realize the application of the proposed method through Matlab program and promote the domestic application of Matlab software.