Best Points Selection Procedure for Estimating Location and Scatter in Multivate Data with Application to Discriminant Analysisari

Abstract

Multivariate data analysis is rely on the two measures namely location and scatter. The most widely used such estimators; sample mean and covariance matrix are extremely sensitive to outliers, then the results obtained with these estimators are inaccurate. Many robust alternatives are established and perform well while handling the data with outliers. But even still a challenging task while handling the large number of cases and/or variables with reference to the features such as dimensionality of data, heterogeneous of data, computing time, adequacy of the results and applications. This paper provides a procedure for the selection of best data points in order to estimate multivariate location and scatter. The obtained results also compared with the established robust procedures such as various MCD algorithmic techniques and MVE by a real environment. The application aspect of the procedure is also executed in the context of discriminant analysis of multivariate grouped data. The results such as apparent error rate, confusion matrix of classical and various robust discriminant procedures are also provided.