Fig 1.
Research framework for predicting the 28-day strength of cement.
Table 1.
Statistical results of feature variables.
Table 2.
XGBoost meta-learners parameters to be optimized.
Fig 2.
Statistical results of 10-fold CV.
Table 3.
Analysis of 10-fold CV results.
Fig 3.
Correlation between cement 28-day strength input and output.
Table 4.
Paired t-test results of the performance differences between TF-XGBoost and XGBoost.
Fig 4.
Average multi-head attention heatmap of typical samples.
Table 5.
Parameter settings for the meta-learner.
Table 6.
Statistics of 5-fold CV for different meta-learners.
Fig 5.
Training efficiency statistics.
Fig 6.
Results of noise robustness verification.
Fig 7.
Results of feature missing robustness verification.
Fig 8.
Statistics of the 90% prediction interval results for typical types of cement.
Table 7.
Model optimization results.
Fig 9.
Results of 25 MC-CV runs on the training and validation sets.
Fig 10.
Performance of different models on the test set.
Table 8.
Ablation experiment evaluation metric comparison.
Fig 11.
Shap analysis results.
Fig 12.
SHAP dependency plot of 3-day compressive strength and other parameters.