Comparing clustering and metaclustering algorithms

Abstract

In this paper, the accuracies of four meta-clustering algorithms and five different base-clustering algorithms are compared. These algorithms come from different knowledge areas such as statistics, neural networks and machine learning. The main advantages of these algorithms are their adaptiveness to some specific datasets. The ensembles are based on bagging, voting and graph partitioning. These ensembles use relabeling techniques to find their accuracy. Mutual information and misclassification error rate are used to compare the performance of these algorithms. One artificial and three real-world datasets available at the UCI Machine Learning Repository are used in the experiments. © 2011 Springer-Verlag.