Majority voting based on q-Gaussian activation function circular extreme learning machine

This paper presents a new approach of Extreme Learning Machine (ELM) ensembles that use majority voting with the q-Gaussian Activation function Circular Extreme Learning Machine (QCELM) to make the final decision for classification problems. For each QCELM is work on the CELM using q-Gaussian activation functions based on Tsallis distribution that varies the different parameter q values (called the entropic index) and selects the best parameter q to take the highest accuracy; As a result, QCELM is more accurate than traditional and applied ELMs. In the experiment, the results of the simulation on 46 datasets from UCI repository shows that Majority Voting based on q-Gaussian Circular Extreme Learning Machine (V-QCELM) outperforms several Extreme Learning Machine ensembles. Wilcoxon signed rank test is used to confirm statistical differences between V-QCELM and the compared ELM ensembles that can indicate V-QCELM is the higher-performance ensemble of Extreme Learning Machines.