Benchmarking taxonomy for 1D clustering algorithms

Abstract

Clustering has been a very active research topic in pattern recognition, and many algorithms and validity indices were proposed. There has been a long debate between fuzzy and crisp clustering and validity indices were proposed as a measure of the correctness of the clustering results. Nevertheless, these indices only verify if the clustering results fit the model they represent and give no information about the true classification of observations. In this paper, we propose a taxonomy to evaluate the performance of clustering algorithms and the subsequent validity indices. The ground-truth data is generated in a way that both the number of clusters and the inter clusters overlap rate are known.