Novel Feature Selection Method for Nonlinear Support Vector Regression

Table 3

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

Data sets	Regressor	NMSE		RMSE	Precision	Recall	CPU S.
Type A	FS-NSVR	0.707	0.146	0.146	22.22	100	0.148
	L1-SVR	1.032	0.031	0.177	16.67	10	2.107
	Lp-SVR	1.028	0.028	0.176	—	0	2.701
	L1-LSSVR	1.028	0.031	0.176	0	0	0.004
Type B	FS-NSVR	0.032	0.884	0.069	34.48	100	0.071
	L1-SVR	1.068	0.074	0.396	1.26	10	1.379
	Lp-SVR	1.077	0.078	0.397	—	0	2.702
	L1-LSSVR	1.001	0.001	0.383	—	0	0.004
Type C	FS-NSVR	0.006	0.885	0.324	18.51	100	0.076
	L1-SVR	1.003	0.009	4.186	0	0	1.229
	Lp-SVR	1.011	0.011	4.201	—	0	2.584
	L1-LSSVR	0.997	0.001	4.174	0	0	0.005

Bold values refer to the best performing regressors under each criterion.