Ensemble Learning Based on Parametric Triangular Norms

Abstract: Along with the increase of data usage in actual applications, it has become an important issue in ensemble learning to improve the ability for data analysis and processing. In order to improve the learning precision and get more accurate classification and projections in practical problems, triangular norms are introduced into the ensemble learning system. Triangular norms can improve the generalization capability of machine learning models. This paper investigates the effect of applying six different parametric triangular norms in ensemble learning system. Firstly, a new combination model for ensemble learning was constructed. Then, six different parametric triangular norms were used respectively as the combination rule of the new model. Finally, genetic algorithm was used as the parameter estimation module of the new rules. There are two kinds of experiments were conducted, they are the classification experiments and the prediction experiments. The classification experiments were conducted on seven different datasets from the University of California, Irvine machine learning repository (UCI). The prediction experiments were conducted on five samples with actual values. The experimental results show that choosing the appropriate combination rule may lead to higher accuracy than the single classifiers or the single prediction models. The improved performance is produced by employing the Yager operator, Aczel-Alsina operator and Schweizer-Sklar operator, and the Yager operator is most suitable for the combination rule of ensemble learning system.