Conclusions from the studies on hybrid ensembles seem to indicate that they will produce
estimators with differing specialities and accuracies in different regions of the space—it
seems sensible that two systems which represent a problem and search the space in radically
different ways will show different strengths, and therefore different patterns of generalisation.
This specialisation implies that with hybrid ensembles, selection of a single estimator rather
than fusion of the outputs of all estimators may be more effective. The dynamic selection