q-Gaussian activation function Circular Extreme learning machine for classification problems

Circular Extreme learning machine (C-ELM) is the one extended Extreme Learning Machine (ELM). It has the same structure as ELM and Circular Back Propagation (CBP) to make it can mapping both linear and circular decision boundaries. The activation function is the one main point to define the decided boundaries. Many activation functions have been proposed. This paper proposes q-Gaussian activation function based on Tsallis statistics that is the one choice to make flexibility decision boundaries with different Gaussian shapes in the Circular Extreme Learning Machine. It calls q-Gaussian activation function Circular Extreme Learning Machine (QC-ELM) that added and varied parameter q (the entropic index) to improve C-ELM's performance. From experimental results of QC-ELM compared with many other methods tested on UCI datasets shows that the QC-ELM has better results in accuracy than many the compared classical methods.