An experimental validation of some indexes of fuzzy clustering similarity

Measuring the similarity between clusterings is a classic problem with several proposed solutions. In this work we focus on measures based on co-association of data pairs and perform some experiments to investigate whether specificities can be highlighted in their behaviour. A unified formalism is used, which allows easy generalization of several indexes to a fuzzy setting. A selection of indexes is presented, and experiments investigate simplified cases and a paradigmatic real-world case, as an illustration of application. © 2009 Springer Berlin Heidelberg.