A Comparative Analysis Between Crisp and Fuzzy Data Clustering Approaches for Traditional and Bioinspired Algorithms

Abstract

AbstractPartitional data clustering algorithms produce a relationship matrix between data and clusters, named membership matrix, which clusters can be treated as mutually exclusive (crisp) or not (fuzzy), according to data clustering approach used. Moreover, a partition obtained by a crisp algorithm can be fuzzified and so, the relationship between data and clusters be relaxed, such as in fuzzy data clustering approach. However, algorithms have your own heuristic and, iteratively, the behavior of a crisp algorithm can be different of that respective fuzzy version and, in addition, fuzzifying a crisp partition can produce different result in relation to crisp and fuzzy clustering algorithms. Therefore, this paper proposes a comparative analysis among results produced by fuzzy data clustering algorithms and their respective crisp versions and fuzzified partitions. The proposal is identify whether a fuzzy clustering algorithm can be replaced by its respective crisp with fuzzified partition, in terms of result quality. The experiments were performed to two traditional partitional algorithms and two bioinspired algorithms.KeywordsFuzzy data clusteringFCMk-MeansParticle Swarm Clustering