On ensembles of biclusters generated by NichePSO

Ensemble methods combine multiple models into a single framework for coping better with Machine Learning tasks. Recently, the well-known Bagging approach was adapted to solve biclustering problems, where the objective is to find large sub-groups of samples and attributes of the data matrix with the samples showing high correlation over the attributes. In this paper, aiming at the generation of more diverse and high-quality biclusters to be fused through an ensemble perspective, we have adopted a well-known multimodal Particle Swarm Optimization algorithm, namely NichePSO. In particular, the study brings a preliminary comparative assessment of the biclustering results delivered by NichePSO operating alone and by two ensemble settings (one of which is Bagging) operating on the biclusters produced by NichePSO. The assessment was done based on bioinformatics and collaborative filtering datasets, and the results achieved so far reveal the usefulness of ensembling the repertory of biclusters produced by NichePSO.