Prediction of Soil Available Boron Content in Visible-Near-Infrared Hyperspectral Based on Different Preprocessing Transformations and Characteristic Wavelengths Modeling
Table 6
The performance and parameters of the best models.
Regression model
Pretreatment method
Test
Test RMSE
RPD level
RPIQ
Parameters
Elastic net
DT
0.75
0.09
A
1.45
Alpha = 2 10−5, L1 = 0.01
Lasso
SG
0.72
0.12
A
1.30
Alpha = 0.0001
Ridge
LG
0.77
0.08
A
1.56
Alpha = 0.0005
BPNN
MSC
0.81
0.37
A
1.55
[16, 8, 4, 1]
SVM_Linear
LG
0.76
0.08
A
1.50
n_components = 3500
SVM_RBF
DT
0.82
0.04
A
2.15
C = 200000, gamma = 1
SVM_Sigmoid
LG
0.76
0.10
A
1.37
Gammas = 5 10−5, C = 6200000, coef = 0
PLS_Linear
SG + LG
0.78
0.07
A
1.66
n_components = 14
PLS_RBF
SG + SNV + DT
0.80
0.04
A
2.12
n_components = 14, gamma = 0.05
PLS_Sigmoid
SG_MC
0.77
0.07
A
1.58
n_components = 15, gamma = 0.002, coef = 0
Note. Bold indicates that the prediction accuracy of the model is good.
