. 2020 May 27;10(34):19852–19860. doi: 10.1039/d0ra00061b

Feature selection and independent prediction accuracy of SVR model.

Methods	Feature number	Dataset 1		Feature number	Dataset 2		Feature number	Dataset 3
Methods	Feature number	MSE	R ²	Feature number	MSE	R ²	Feature number	MSE	R ²
All	1219	0.1066	0.7793	1323	0.1740	0.8389	1360	0.1709	0.7468
R ^a	19	0.0626	0.8686	65	0.0489	0.9658	91	0.3431	0.4541
R ^b	20	0.0994	0.8121	18	0.0477	0.9503	37	0.3655	0.4445
dCor^a	49	0.0948	0.7873	88	0.0283	0.9733	100	0.2358	0.6212
dCor^b	15	0.0701	0.8368	42	0.0229	0.9767	25	0.1640	0.7518
Chi-MIC^a	86	0.0985	0.7842	61	0.0561	0.9467	82	0.2488	0.5975
Chi-MIC^b	27	0.1387	0.7029	34	0.0791	0.9716	15	0.4184	0.3631
mRMR^a	15	0.1339	0.7180	98	0.1088	0.8876	70	0.1686	0.7503
mRMR^b	13	0.1291	0.7188	26	0.1139	0.8578	11	0.2968	0.5607
Chi-MIC-share	15	0.0280	0.9590	27	0.0226	0.9750	22	0.0454	0.9367

Forward selection method without culling feature.

Forward selection method with culling feature.