An Improved Kernel Based Extreme Learning Machine for Robot Execution Failures

Table 4

Comparison of performance by AKELM and KELM learning algorithms for the regression problems.

Algorithms with different kernel functions	Box and Jenkins gas furnace data			Auto-Mpg
	Training error	Testing error	Training time (seconds)	Training error	Testing; error	Training time (seconds)
KELM (parameters = 1, Gaussian)	0.0120	0.0188	0.0394	0.0529	0.0599	0.1213
KELM (parameters = 1, tangent)	0.0627	0.0655	0.0116	0.6680	0.7756	0.0346
KELM (parameters = 1, wavelet)	0.0121	0.0206	0.0177	0.0509	0.0597	0.0415
KELM (parameters = 10, Gaussian)	0.0183	0.0213	0.0149	0.0685	0.0732	0.0286
KELM (parameters = 10, tangent)	0.2245	0.1986	0.0044	0.2071	0.2085	0.0261
KELM (parameters = 10, wavelet)	0.0306	0.0382	0.0101	0.0662	0.0712	0.0360
AKELM (Gaussian)	0.0133	0.0183	26.1250	0.0503	0.0597	74.7656
AKELM (tangent)	0.0223	0.0242	25.2500	0.0735	0.0735	73.8906
AKELM (wavelet)	0.0133	0.0183	28.3906	0.0502	0.0597	84.9688