Ensemble approach for clustering of interval-valued symbolic data

Abstract

Ensemble approach has been applied with a success to regression and discrimination tasks [see for example Gatnar 2008]. Nevertheless, the idea of ensemble approach, that is combining (aggregating) the results of many base models, can be applied to cluster analysis of symbolic data. The aim of the article is to present suitable ensemble clustering based on symbolic data. The empirical part of the paper presents results simulation studies (based on artificial data sets with known cluster structure) of ensemble clustering based on co-occurrence matrix for symbolic interval-valued data, compared with single clustering method. The results are compared according to corrected Rand index.