Ensemble Learning Based on Group Decision Making Using Projection Method

Ensemble learning (EL) is a machine learning paradigm which aims to improve the generalization performance of models. In this paper, EL based on group decision making (GDM) using projection method is proposed, which takes base learners as decision makers and regards combination of EL as the process of GDM. Specifically, after the base learners have been trained on the training set, their performance on the validation set will be taken as the prior information of learners. Next, several attributes are derived and calculated in the test set, which constitutes multiple attribute group decision making formulation. Then GDM based on projection method is applied to determine the weights of base learners. Finally, the integrated results are obtained and the highest is taken as predictive label. The experimental results in a real industrial dataset demonstrate the effectiveness and superiority of the proposed EL method by comparing with other typical EL methods.