A Preliminary Study of Influential Observation Measurement in K-Means Clustering Procedure

Abstract

An influential observation measurement procedure was developed in the context of cluster analysis, and in particular, k-means partitioning procedure (using the Jancey and the Forgy algorithms) adapting the idea of “leave-one-out” method. A computer program called DETLIE was developed and then applied to both synthetic data and empirical data sets to explore the applicability of the developed procedure of influential observation measurement. The results are promising given the exploratory nature of the current study. Directions for future research were also discussed.