Cluster ensemble has proved to be a good alternative when facing cluster analysis problems. It consists of generating a set of clusterings from the same dataset and combining them into a final clustering. The goal of this combination process is to improve the quality of individual data clusterings. Due to the increasing appearance of new methods, their promising results and the great number of applications, we consider that it is necessary to make a critical analysis of the existing techniques and future projections. This paper presents an overview of clustering ensemble methods that can be very useful for the community of clustering practitioners. The characteristics of several methods are discussed, which may help in the selection of the most appropriate one to solve a problem at hand. We also present a taxonomy of these techniques and illustrate some important applications.

No abstract

No abstract

Cluster ensemble is a good alternative to face the problem of data clustering. Some studies based on mathematical models have shown that cluster ensemble methods lead to an effective improvement of the results of the standard clustering algorithms. In this paper, we focus on this problem, proposing a new approach to solve it, by adding a new step into the usual cluster ensemble methodology. Representing partitions by graphs and a new kernel function to measure the similarity between partitions are other proposals for this work. Experiments with synthetic and real databases show the suitability and effectiveness of our method.

The analysis of human brain connectivity networks has become an increasingly prevalent task in neuroimaging. A few recent studies have shown the possibility of decoding brain states based on brain graph classification. Graph kernels have emerged as a powerful tool for graph comparison that allows the direct use of machine learning classifiers on brain graph collections. They allow classifying graphs with different number of nodes and therefore the inter-subject analysis without any kind of previous alignment of individual subject's data. Using whole-brain fMRI data, in this paper we present a method based on graph kernels that provides above-chance accuracy results for the inter-subject discrimination of two different types of auditory stimuli. We focus our research on determining whether this method is sensitive to the relational information in the data. Indeed, we show that the discriminative information is not only coming from topological features of the graphs like node degree distribution, but also from more complex relational patterns in the neighborhood of each node. Moreover, we investigate the suitability of two different graph representation methods, both based on data-driven parcellation techniques. Finally, we study the influence of noisy connections in our graphs and provide a way to alleviate this problem.