Robust Extreme Learning Machine Using New Activation and Loss Functions Based on M-Estimation for Regression and Classification

Table 6

Experimental results on real–world dataset abalone with 0%–20% Outlier's levels.

Wang et al. [35] used sigmoid activation function in existing ELM and its variants
ELM		WELM	ORELM	IRRELM	IRRELM
Outliers	(RMSE, SD)	(RMSE, SD)	(RMSE, SD)	(RMSE, SD)	(RMSE, SD)
0%	2.1382, 0.0692	2.1532, 0.0737	2.1909, 0.0576	2.2106, 0.0646	2.1350, 0.0625
5%	2.3182, 0.0533	2.1557, 0.0609	2.1712, 0.0589	2.1928, 0.0632	2.1455, 0.0552
10%	2.6679, 0.0533	2.1603, 0.0597	2.1667, 0.082	2.1724, 0.0579	2.1499, 0.0621
15%	3.1917, 0.0763	2.1701, 0.0909	2.1713, 0.0820	2.1492, 0.0603	2.1508, 0.0643
20%	3.2161, 0.0976	2.2297, 0.0636	2.2168, 0.0741	2.1694, 0.0755	2.1633, 0.0716