A Method to Improve the Analysis of Cluster Ensembles

Abstract Clustering is fundamental to understand the structure of data. In the past decade the cluster ensemble problem has been introduced, which combines a set of partitions (an ensemble) of the data to obtain a single consensus solution that outperforms all the ensemble members. However, there is disagreement about which are the best ensemble characteristics to obtain a good performance: some authors have suggested that highly different partitions within the ensemble are beneficial for the final performance, whereas others have stated that medium diversity among them is better. While there are several measures to quantify the diversity, a better method to analyze the best ensemble characteristics is necessary. This paper introduces a new ensemble generation strategy and a method to make slight changes in its structure. Experimental results on six datasets suggest that this is an important step towards a more systematic approach to analyze the impact of the ensemble characteristics on the overall consensus performance.