Discriminant Analysis by Projection Pursuit

Abstract

A non-parametric discriminant analysis (projection pursuit by principal component analysis) is discussed and used to compare three robust linear discriminant functions that are based on high breakdown point (of location and covariance matrix ) estimators. The major part of this paper deals with practical application of projection pursuit by principal component. In this study 10 simulated data sets that are binomially distributed and a real data set on the yield of two different progenies of palm tree were used for comparisons. From the findings we concluded that the non-parametric procedure (projection pursuit by principal component) have the highest predictive power among other procedures we considered. S-estimator performed better than the other two estimators when real data is considered, while MCD estimator performed better than MWCD estimator.