Adaptive Online Sequential ELM for Concept Drift Tackling

Table 9

Performance comparison of AOS-ELM and nonadaptive OS-ELM to validation data set in %.

(a) Benchmark result, nonadaptive OS-ELM

Concept	Parameters	Testing accuracy	Cohen’s kappa
	OS-ELM	50.87 ± 0.01	45.81 (0.54)
	OS-ELM	94.58 ± 0.05	94.03 (0.24)

(b) Performance result on the drift event, AOS-ELM

Drift	Parameters	Testing accuracy	Cohen’s kappa
	AOS-ELM1	92.18 ± 2.73	91.38 (0.29)
	AOS-ELM2	94.64 ± 0.06	94.10 (0.24)
End of full	AOS-ELM1	93.45 ± 1.18	92.78 (0.27)
End of full	AOS-ELM2	94.57 ± 0.11	94.02 (0.25)