Canonical Discriminant Analysis: Comparison of Resampling Methods. and Convex-hull Approximation

Abstract

Abstract In order to characterize the predictive power of multivariate data-analytic methods like Principal Components Regression, Partial Least Squares analysis and Canonical Discriminant Analysis, it is customary to use resampling methods like the jackknife or the bootstrap (see e.g. Krzanowski and Radley, 1989; Weihs and Schmidli, 1991). Predictive distributions are es pecially valuable in model determination, i.e. in the examination of model adequacy and in model selection. Also in this field, a cross-validation view point has been advocated by Gelfand et al (1991).