다목적 다변량 자료분석을 위한 변수선택

Abstract

Recently we frequently analyze multivariate data with quite large number of vari-ables. In such data sets, virtually duplicated variables may exist simultaneously eventhough they are conceptually distinguishable. Duplicate variables may cause problemssuch as the distortion of principal axes in principal component analysis and factoranalysis and the distortion of the distances between observations,i.e. the input forcluster analysis. Also in supervised learning or regression analysis, duplicated explana-tory variables often cause the instability of tted models. Since real data analyses areaimed often at multiple purposes, it is necessary to reduce the number of variables toa parsimonious level.The aim of this paper is to propose a practical algorithm for selection of a subsetof variables from a given set ofp input variables, by the criterion of minimum traceof partial variances of unselected variables unexplained by selected variables. Theusefulness of proposed method is demonstrated in visualizing the relationship betweenselected and unselected variables, in building a predictive model with very large numberof independent variables, and in reducing the number of variables and purging/mergingcategories in categorical data.