A Comparison of Sparse Partial Least Squares and Elastic Net in Wavelength Selection on NIR Spectroscopy Data

Table 1

Comparison of Enet and SPLS under combinations of different and based on the 100 replications. is the relative prediction error, “C” and “IC” are the number of predictors that are correctly and incorrectly selected into the model, respectively.

	Model		C	IC		C	IC		C	IC
100/25/6	Truth		6.000	0.000		6.000	0.000		6.000	0.000
	Enet	0.737	5.930	0.160	0.590	5.940	0.130	0.128	6.000	1.150
	SPLS	0.180	6.000	1.810	0.140	6.000	1.630	0.172	5.950	2.610
200/37/12	Truth		12.000	0.000		12.000	0.000		12.000	0.000
	Enet	2.796	11.440	0.020	5.500	10.320	0.000	7.937	9.820	0.000
	SPLS	0.121	11.990	1.440	0.140	11.990	1.900	0.230	11.990	6.080
400/55/18	Truth		18.000	0.000		18.000	0.000		18.000	0.000
	Enet	4.516	17.290	0.030	12.770	14.780	0.000	16.153	12.920	0.000
	SPLS	0.097	18.000	2.020	0.110	18.000	2.830	0.204	17.980	6.710
100/120/6	Truth		6.000	0.000		6.000	0.000		6.000	0.000
	Enet	0.709	5.930	0.200	5.530	6.000	0.700	5.555	6.000	1.970
	SPLS	0.124	5.990	0.480	0.190	5.990	0.760	0.246	5.970	2.990
100/300/15	Truth		15.000	0.000		15.000	0.000		15.000	0.000
	Enet	5.837	13.550	2.420	10.640	12.090	0.020	8.191	12.130	0.010
	SPLS	0.434	14.930	4.460	0.370	14.960	2.820	0.679	14.850	4.660
100/800/35	Truth		35.000	0.000		35.000	0.000		35.000	0.000
	Enet	33.360	25.300	18.000	50.732	22.670	0.860	6511.400	32.480	0.890
	SPLS	2.440	28.000	33.670	1.620	32.600	17.200	1.730	34.590	7.030