Comparative performance evaluation of global-local hybrid ensemble

Article type: Research Article

Authors: Baumgartner, Dustin | Serpen, Gursel

Affiliations: Electrical Engineering and Computer Science, University of Toledo, Toledo OH, 43606, USA

Abstract: This paper presents a comprehensive simulation study which aims to profile the performance capabilities of the global-local hybrid ensemble in comparison with leading ensemble classifiers as reported in recent studies in the literature. The global-local hybrid ensemble is implemented with decision tree (global) and nearest-neighbor (local) base learners and its accuracy performance is compared, on 46 benchmark datasets from the UCI machine learning repository, to those of other ensembles from six prominent studies in the literature. Through statistical significance testing, it is shown that the global-local hybrid ensemble is a robust classifier design: over a larger spectrum of data domains, it performs competitively with other leading ensembles.

Keywords: Hybrid ensemble, classification, global/local learning, heterogeneous/homogeneous diversity, nonparametric comparison

DOI: 10.3233/HIS-2011-0122

Journal: International Journal of Hybrid Intelligent Systems, vol. 8, no. 2, pp. 59-70, 2011