Performance of Eight Dissimilarity Coefficients to Cluster a Compositional Data Set
Concerned with the problem of clustering a compositional data set consisting of vectors of positive components subject to a unit-sum constraint, as a first step we looked for an appropriate dissimilarity coefficient or distance between two compositions. In this paper we selected eight different dissimilarity measures, and their performance was evaluated by means of graphics and cluster validity coefficients of six clustering methods applied to three compositional data sets. Almost recent criteria for measures of compositional difference are also tested for those measures emerging as the best to cluster compositions.
KeywordsCompositional Data Single Linkage Dissimilarity Measure Ternary Diagram Cluster Validity
Unable to display preview. Download preview PDF.