Brkga Algorithm Applied to the Cluster Ensemble Problem

Clustering algorithms are used to partition datasets associated with variousreal-world applications. However, in addition to the adopted algorithm, thequality of the obtained partition depends on the data distribution. Consequently, applying a single algorithm can, in many cases, lead to producing a poor-quality partition (measured by some validation index). Considering this fact, Cluster Ensemble is an alternative to produce a good quality partition,since combines different dataset partitions into a single consensus partition.According to the literature, in general, the consensus partition is less sensi-tive to noise in the data and presents higher quality when compared to thatproduced by a single algorithm.In order to produce a good quality consensus partition, this work proposesa Cluster Ensemble algorithm that combines: (i) three different strategies forproducing base partitions; (ii) the BRKGA metaheuristic, and (iii) the useof the mean silhouette index. To evaluate the new algorithm, computationalexperiments were carried out with 20 databases and several algorithms fromthe literature, where two well-known external validation indices (NMI andAR) and hypothesis tests were applied. As a result, it was observed that theproposed algorithm presented good-quality solutions, in the three strategies,for different datasets compared to the main algorithms in the literature.