Stochastic ensemble pruning method via simulated quenching walking

摘要: Inspired by an upward stochastic walking idea, a new ensemble pruning method called simulated quenching (SQWALKING) is developed in this paper. The rationale behind to give values movements as well accept unvalued solutions during the investigation of search spaces. SQWALKING incorporates and forward selection methods choose models through using probabilistic steps. Two versions are introduced based on two different evaluation measures; SQWALKINGA that accuracy measure SQWALKINGH human-like foresight measure. main objective construct proper architecture pruning, which independent construction combination phases. Extensive comparisons between proposed competitors terms heterogeneous homogeneous ensembles performed ten datasets. show can lead respectively 5.13% 4.22% average improvement. One reason for these promising results phase takes additional time find best compared rivals. Finally, SQWALKINGs also evaluated real-world dataset.