Cost-sensitive Hierarchical Clustering for Dynamic Classifier Selection

Meinolf Sellmann, Tapan Shah

Abstract: Given an ensemble of classifiers, dynamic classifier selection (DCS) selects one classifier depending on the particular input vector that we get to classify. DCS is a special case of algorithm selection (AS) where we can choose from multiple different algorithms to process a given input. We investigate if cost-sensitive hierarchical clustering (CSHC), a method originally developed for AS, is suited for DCS. We tailor CSHC for the special case of choosing a classification algorithm and compare with state-of-the-art DCS methods. We then show how the new methodology can be used for stacking. Experimental results show that CSHC-based DCS outperforms the best methods to date.