Novel Feature Selection Method for Nonlinear Support Vector Regression

Table 5

Comparison results of FS-NSVR, L₁-SVR, L_p-SVR, and L₁-LSSVR for real-world datasets.

Data sets	Regressor	(%)	NMSE	RMSE		CPU S.
Student	FS-NSVR	6.250	0.094	0.037	0.913	0.070
	L1-SVR	72.917	0.062	0.030	0.956	1.118
	Lp-SVR	62.500	0.065	0.030	0.980	0.400
	L1-LSSVR	52.083	0.062	0.025	0.960	0.001
Traffic	FS-NSVR	18.519	0.905	0.946	0.102	2.007
	L1-SVR	85.185	0.963	0.976	0.252	2.259
	Lp-SVR	88.889	1.083	1.035	0.530	0.457
	L1-LSSVR	44.444	0.952	1.602	0.057	0.001
CNN	FS-NSVR	7.407	0.001	0.035	1.000	0.237
	L1-SVR	86.420	0.00	0.011	1.000	1.249
	Lp-SVR	9.877	0.001	0.032	0.981	0.372
	L1-LSSVR	8.642	0.000	0.020	0.983	0.001
Community	FS-NSVR	21.569	0.379	0.138	0.918	1.108
	L1-SVR	66.667	0.331	0.129	0.656	2.047
	Lp-SVR	37.255	0.337	0.130	0.637	1.822
	L1-LSSVR	74.510	0.340	0.130	0.731	0.001
Pakinson	FS-NSVR	23.077	0.862	0.446	0.415	1.176
	L1-SVR	42.308	1.093	0.502	0.319	1.276
	Lp-SVR	88.462	1.229	0.533	0.430	0.570
	L1-LSSVR	76.923	1.021	0.372	0.197	0.001