Visual Exploration Tools for Ensemble Clustering Analysis
Sonia Fiol-González, Cassio Almeida, Ariane Nunes Rodrigues, Simone Diniz Junqueira Barbosa, Hélio Lopes · 2019
Uncertainty Analysis is essential to support decisions, and it has been gaining attention in both visualization and machine learning communities —in the latter case, mainly because ensemble methods are becoming a robust approach in several applications. In particular, for unsupervised learning, there are several ensemble clustering methods that generate a co-association matrix, i.e., a matrix whose element (i, j) represents the estimated probability that the given sample pair is on the same cluster. This work studies the following decision problem: “Given a similarity function, which groups of elements of a set form robust clusters?” Robust here means that all elements of each cluster are connected with a probability within a given interval. Our main contribution is a prototype that helps decision makers, through visual exploration, to have insights to solve this task. To do so, we provide visual tools for ensemble clustering analysis. Such tools are grounded in the co-association matrix generated by the ensemble. With these tools we are better equipped to recommend the group of elements that form each cluster, considering the uncertainty generated by ensemble clustering methods.