BEMI Bicluster Ensemble Using Mutual Information

Biclustering solutions generally depend upon various parameters like number of biclusters and random initialisations. Ensemble techniques have been used to eliminate the impact of such parameters on the output. In this paper, we present a novel ensemble technique for biclustering solutions using mutual information. Unlike the existing approaches, the proposed technique does not require the biclusters to be aligned. As a result, it does away with the requirement that all the biclustering solutions generate the same number of biclusters. Moreover, most of the existing approaches require the user to specify the number of output biclusters. Our approach determines the number of well separated biclusters from the input solutions itself. Experiments performed on synthetic and real datasets show that our approach improves upon the biclustering error over the input solutions as well as the ensemble techniques of hanczar et al.