Table 1.
Comparison of the results using six different variable selection methods.
Fig 1.
Plot of coefficients obtained by PLS, Enet and Enet-BETA.
(a) This figure shows the regression coefficients of 100 variables obtained by PLS, and it means that PLS selects all the process variables into the regression model. (b) This figure shows the regression coefficients of 100 variables obtained by Enet, and we can see that part of coefficients are shrank to be zero, which means that Enet selects part of the process variables into regression model. (c) This figure shows the regression coefficients of 100 variables obtained by Enet-BETA, and it’s clear that only a small part of variables are selected into regression model.
Fig 2.
Plot of predicted vs. measured value of Enet-BETA.
The scatter plot depicts the prediction accuracy of the built model. The x axis represents the measured value of the percentage of protein concentration and the y axis represents the predicted value by the regression model.
Fig 3.
Plot of predicted vs. measured value of Enet-BETA.
This plot depicts the spectra of different LGA solution concentrations at 9.0, 15.0, 21.0, 27.0, 33.0, 39.0 g/L, and it’s clear that there exist high-colinearity between process variables.
Table 2.
Comparison of the results using six different variable selection methods.
Fig 4.
Plot of coefficients obtained by PLS, Enet and Enet-BETA.
(a) This figure shows the regression coefficients of 216 variables obtained by PLS, and it means that PLS selects all the process variables into the regression model. (b) This figure shows the regression coefficients of 216 variables obtained by Enet, and we can see that part of coefficients are shrank to be zero, which means that Enet selects part of the process variables into regression model. (c) This figure shows the regression coefficients of 216 variables obtained by Enet-BETA, and it’s clear that only a small part of variables are selected into regression model.
Fig 5.
Plot of predicted vs. measured value of test data.
The scatter plot depicts the prediction accuracy for the first test data with. The x axis represents the measured value of the percentage of protein concentration and the y axis represents the predicted value by the regression model.