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

Table 7

Percent accuracy with SD of different existing and proposed activation functions in basic ELM.

Af	ELM 50 IRIS	ELM 300 SatImage	ELM 1000 E-mail spam filtering
Sigmoid	95.67, 4.83	87.5, 0.518874500	88.00000, 1.387233
Tan-sigmoid	96.00, 2.11	87.7857, 0.6992932	88.00000, 1.732051
Sine	95.33333, 4.216370	88.5714, 0.9376145	88.66667, 1.527525
Cosine	98.00000, 2.810913	88.3571, 0.6333237	88.66667, 1.527525
Bentidle	97.00000, 3.314763	86.21429, 0.8418974	90.33333, 0.5773503
RAF	96.00000, 3.784308	88.92857, 0.8287419	76.33333, 0.5773503
Proposed-1	97.66667, 2.249829	88.07143, 0.6157279	88.33333, 1.527525
Proposed-2	95.66667, 3.531166	88.21429, 0.6112498	88.66667, 1.154701
Proposed-3	85.66667, 6.675920	75.00, 0.6793662000	88.66667, 1.527525
Proposed-4	96.00000, 4.097575	89.85714, 0.8644378	88.66667, 1.527525
Proposed-5	96.66667, 2.721655	87.78571, 0.5789342	89.00000, 1.732051
Proposed-6	97.00000, 4.830459	86.85714, 0.7449463	89.00000, 1.732051
Proposed-7	96.33333, 1.892154	60.0, 5.7912400000	70.66667, 1.154701
Proposed-8	98.33333, 2.357023	88.07143, 0.9168748	93.66667, 0.5773503
Proposed-9	95.66667, 2.744242	87.64286, 0.7449463	93.66667, 0.5773503
Proposed-10	96.66667, 3.513642	77.78571, 1.5281250	71.66667, 0.5773503
Proposed-11	95.33333, 4.499657	88.28571, 0.6112498	71.66667, 0.5773503