Pivotal seeding for K-means based on clustering ensembles

Despite its large use, one major limitation of K-means algorithm is the impact of the initial seeding on the final partition. We propose a modified version, using the information contained in a co-association matrix obtained from clustering ensembles; such matrix is given as input for a set of pivotal methods, implemented in the pivmet R package, used to perform a pivot-based initialization step. Preliminary results concerning the comparison with the classical approach and other clustering methods are discussed.