PAM modification using trimmed K-median based on TCLUST cluster analysis

Abstract

This paper will discuss the TCLUST algorithm using restriction of constrains to scatter matrices. We are discussing among three constrains eigenvalue, matrix determinant and same sized cluster (sigma) that affect the shape of clusters. Trimming process using TCLUST is made to detect the best proportion of contaminated data and the best number of clusters to be used in the
next step. Based on prior knowledge of TCLUST we are using the PAM to determine the best mediod that shape the data. The results are discussed between the three types of constraints. At the end of this paper we compared the TLUCT based on trimmed k-means method with modified PAM based on trimmed k-median method.