Dynamic Classifier Ensemble Selection Based on GMDH

Dynamic classifier selection (DCS) plays a strategic role in the field of multiple classifiers system. This article introduces group method of data handing (GMDH) theory to DCS, and presents a novel strategy GAES for adaptive classifier ensemble selection first. Then it extends this algorithm and proposes dynamic classifier ensemble selection based on GMDH (GDES). For each test pattern, GDES is able to select an appropriate ensemble from the classifier pool adaptively, determine the combination weights among base classifiers, and complete the combination process automatically. We experimentally test GDES over 6 UCI data sets. The results clearly show that, GDES outperforms the fusion method MAJ (Xu et al., 1992) and also performs slightly better than DCS-LCA (Woods et al., 1997) and KNORA (Ko et al., 2008).