3.3.2 Model Development and Evaluation
In this study, four machine learning algorithms were used to establish
the meta-model, and the features of the four meta-models were fitted by
using the Stacking algorithm. After hyperparameter optimization, the
RMSE values of the obtained four meta-models are all lower than 0.32,
and the prediction ability of the models is good. This study uses four
machine learning algorithms to establish the basic model, and uses the
Stacking algorithm to fit the characteristics of these four basic
models. After optimization of hyperparameters, the RMSE values of the
four models were all lower than 0.43, and the prediction ability of the
models was good (Figure 9A). These models are then fitted using the
stacking model and regularized. The results show that when thepenalty parameter is 0.0173, the mixture value is 1, that
is, the meta-model is completely used for prediction under thispenalty parameter value, and the corresponding RMSE is
0.208, and R2 is 0.954, indicating that the
model prediction performance is good (Figure 9B). At the same time, theRMSE values of the meta-model in the training set and test set
are 0.193 and 0.289, the R2 is 0.962 and 0,928,
and the MAE is 0.068 and 0.116, respectively, and the model
performance is good (Figure 9C, Figure 9D).