2-Degree Polynomial Circular Extreme Learning Machine for Classification Problem

Abstract

The Circular Extreme Learning Machine (CELM) is the combination of Extreme Learning Machine (ELM) and Circular Back Propagation (CBP). CELM uses the structure same as ELM to get speedy in the training process and gets benefit from CBP architecture by using the 2-degree polynomial to make the better performance. However, the using 2-degree polynomial of CELM can create the decided boundary shape similar as RBF that may be limited the distance, between the center point and the data by the sigmoid activation function. So that, a 2-degree Polynomial Circular Extreme Learning Machine (PCELM) is proposed to tackle this problem. Our experiment showed that PCELM outperformed the original ELM and the traditional CELM with several activation functions. Wilcoxon signed rank test was used to compare statistical differences between PCELM and the CELM that can confirm the PCELM can improve the performance of CELM.