Multi-facet determination for clustering with Bayesian networks
Fernando Rodríguez-Sánchez, Pedro Larrañaga Múgica, María Concepción Bielza Lozoya · Hispana · 2017
Real world applications of sectors like industry, healthcare or finance usually generate data of high complexity that can be interpreted from different viewpoints. When clustering this type of data, a single set of clusters may not suffice, hence the necessity of methods that generate multiple clusterings that represent different perspectives. In this paper, we present a novel multi-partition clustering method that returns several interesting and non-redundant solutions, where each of them is a data partition with an associated facet of data. Each of these facets represents a subset of the original attributes that is selected using our information-theoretic criterion UMRMR. Our approach is based on an optimization procedure that takes advantage of the Bayesian network factorization to provide high quality solutions in a fraction of the time.