Using dual evolutionary search to construct decision tree based ensemble classifier

Abstract A typical ensemble learning process typically uses a forward integration mechanism to construct the classifier with large number of base classifiers. Based on this mechanism, it is difficult adjust diversity among classifiers and optimize structure inside since generation has certain amount randomness, which makes performance heavily dependent human design decisions. To address issue, we proposed an automatic construction method based dual-layer evolutionary search includes tree coding-based population binary population. Through collaborative searching between two populations, can be driven by training data update globally. verify effectiveness dual (DEEL), tested 22 classification tasks from 4 repositories. The results show that generate diverse decision while constructing them. Compared 9 competitor algorithms, achieved best 17 test improved average accuracies 0.97–7.65% over second place. In particular, generated excellent structure, involve small trees. That increases transparency ensembles helps perform interpretability analysis