文档介绍：Abstract Dimension reduction one hand,high dimension dataCan notbe used directly insome practical algorithms,while dimension reduction ishelpful to solve theproblem,which iscalled'dimension CUl"se’,byreducing plexity. Then some algorithms theotherhand,hi曲dimension datausually includes agreat dealof’noise’and'redundancy'.Dimension reduction helppeople find interesting data lre in low dimension space forunderstanding research object betterandbetter. PCA and 2DPCA aretwodifferentmethods forreducing dimension On matrixdata. This paper pare PCA and 2DPCA inresearching two kinds ofimportant matrix data:multivariate time seriesdata and high-frequency we can use thetwokinds ofdataforreference topractical application. Inthe classification problem ofmultivariatetime series data,we firstlyreduce dimension andthen classifydata this·purpose,this binations ofPCA and 2DPCA wim Euclidean Distance respectively about classification Euclidean Distance isimpacted by thispaper puts forward anew algorithm,which iscalled”2DPCA's Mahalanobis distancealgorithm ontwo-dimensional principal ”.And this paper pares thenew algorithms to Euclidean Distance algorithm and PCA's Mahalanobis distance algorithmby 5realworld multivariatetime results show that”2DPCA’S Mahalanobis distance algorithm on two-dimensional principal subspace‘。is bination ofdimension reduction method and classification distance. Intheproblem ofstatisticalmodeling on high-frequency financial data,it,s very important to forecast thehigh—frequency financial datawhich containsmany assets has acovariancematrix ofvolatilitiesinevery kind ofcovariance matrixusually has relatively weforecast the volatilitiesdirectly,itwill create agreatnumber firstly we need to 万方数据 reduce dimension and then model low dimension paper make parativ