Bio-inspired algorithms for improving mixed and incomplete data clustering

This article introduces a novel clustering algorithm for dealing with mixed and incomplete data descriptions of instances: The A2FCPAntSA algorithm. Unlike previous proposals, this new algorithm uses groups of instances as initial clusters in an agglomerative clustering scheme. It also merges in a single step, all of the most similar clusters. In addition, the proposal incorporates the use of bio-inspired algorithms to refine the obtained clusters. The numerical experiments carried out over several repository datasets display the superiority of the proposal with respect to other state of the art clustering algorithms, for mixed and incomplete data, by considering several dissimilarity measures and supervised cluster validity indexes.