Abstract:
Triangular norms can improve the generalization capability of pattern classification problems, so they are introduced to improve the performance of ensemble learning. This study pays attention to the different impacts of base classifiers and put forward a new fuzzy combination method for ensemble learning. It is based on parametric triangle norms. Firstly, a new combination model was constructed. The base learners in it are set with different weights. Secondly, a set of fuzzy combination rules were generated according to the parametric triangle norms. Thirdly, genetic algorithm was used as the parameters estimation module of the new fuzzy rules. Finally, experiments were conducted on seven different datasets from the University of California, Irvine machine learning repository (UCI). The experimental results show that the fuzzy rules generated by the new combination method have better performance than the base learners and the fixed combination rules. And when set proper weights to base learners, the new fuzzy rules obtain better performance than the ones set a same weight to base learners.
Subscribe Today 
