A Progressive Combined Variable Selection Method for Near-Infrared Spectral Analysis Based on Three-Step Hybrid Strategy
Table 3
Prediction results of oil content in corn by different variable selection methods.
Variable selection methods
Number of variables
nLV
Rc2
RMSEC (%)
Rp2
RMSEP (%)
None
700
10
0.934
4.77
0.720
9.90
CARS
19
6
0.985 ± 0.003
2.27 ± 0.24
0.933 ± 0.010
3.34 ± 0.35
VCPA
8
6
0.985 ± 0.011
2.29 ± 0.37
0.954 ± 0.004
2.86 ± 1.01
IRIV
34
8
0.956 ± 0.013
3.87 ± 0.76
0.745 ± 0.035
6.69 ± 1.25
SiPLS
140
7
0.993 ± 0.002
1.41 ± 0.47
0.984 ± 0.007
1.68 ± 0.92
IVSO
89
6
0.996 ± 0.002
1.12 ± 0.36
0.977 ± 0.008
2.01 ± 1.14
BOSS
22
8
0.992 ± 0.003
1.69 ± 0.58
0.958 ± 0.008
2.73 ± 0.64
IVSO-BOSS
26
7
0.990 ± 0.002
1.38 ± 0.11
0.991 ± 0.003
1.63 ± 0.28
SiPLS-IVSO
52
8
0.996 ± 0.001
1.02 ± 0.28
0.994 ± 0.001
1.49 ± 0.79
SiPLS-BOSS
21
8
0.996 ± 0.001
1.02 ± 0.38
0.990 ± 0.001
1.83 ± 0.47
CARS-IVSO-BOSS
6
7
0.758 ± 0.021
6.58 ± 0.82
0.613 ± 0.047
12.54 ± 1.02
SiPLS-VCPA-BOSS
9
6
0.992 ± 0.004
1.64 ± 0.34
0.980 ± 0.005
2.30 ± 0.53
SiPLS-IVSO-IRIV
35
5
0.997 ± 0.003
0.93 ± 0.21
0.992 ± 0.007
1.48 ± 0.49
SiPLS-IVSO-BOSS
7
6
0.998 ± 0.001
0.88 ± 0.12
0.993 ± 0.001
1.35 ± 0.29
Notes. nLV – number of latent variables. Rc2 – determination coefficient of calibration. Rp2 – determination coefficient of prediction. RMSEC – root mean square error of calibration. RMSEP – root mean square error of prediction. Models were built by partial least squares method under different variable selection methods and no method. The statistical results were expressed as mean ± standard deviation of 50 runs.
