A Novel Ensemble Method of RVFL For Classification Problem

Ensemble learning methods, which combine several base classifiers, is a common technique to enhance the classification ability of ensemble models in the field of pattern recognition and machine learning. Rotation Forest, an ensemble algorithm, has been used widely in various fields with nice generalization performance. The main idea of Rotation Forest is to animate concurrently both diversity and individual accuracy within the ensemble. On the other hand, random vector functional link (RVFL) neural network, a randomized version of single layer feed-forward neural network (SLFN), is a successful model because of its universal approximation property. In this paper, we propose a novel ensemble method, known as rotated random vector functional link neural network (RoF-RVFL), which combines rotation forest (RoF) and RVFL classifiers. To verify the effectiveness of the proposed RoF- RVFL method, empirical comparisons are carried out among Rotation Forest (RoF), Random Forest (RaF), RVFL and the proposed RoF-RVFL method over 42 UCI benchmark datasets. The experimental results show that the proposed RoF-Rvflmethod is able to generate more robust network with better generalization performance.