Fig 1.
Methodology flowchart of the research.
Fig 2.
Handy specimens and microscopic images of four tested rocks as representative samples.
(Note: SC: Sparry calcite, M: Micrite, C: Cavity, V: Vein, F: Fossil).
Table 1.
Engineering properties of the rocks.
Fig 3.
Distribution curves and histograms of the rock properties in the database.
Fig 4.
Pearson’s correlation coefficient for data.
Table 2.
Descriptive statistics of data set.
Fig 5.
Training phase of the stacking model.
Fig 6.
Predictions process performed by a model in the first level of the stacking ensemble.
Fig 7.
Grid search with 3-fold cross validation.
Fig 8.
Pseudocode of building models in the stacking ensemble.
Table 3.
Parameter setup for each model.
Fig 9.
Effects of the number of neurons in hidden layer and the solver of weight optimization in the MLP on MSE of predicting a) UCS, and b) E.
Fig 10.
Effects of parameters of RF on MSE of forecasting a) UCS, and b) E.
Fig 11.
Effect of parameters of the SVR on MSE of predicting a) UCS, and b) E.
Fig 12.
Effects of parameters of the XGBoost on MSE of predicting a) UCS, and b) E.
Fig 13.
Effects of parameters of a) SVR, and b) MLP as the meta-learner for predicting the UCS, and E, respectively.
Table 4.
Performance results for predicting the UCS (MPa) in the testing phase.
Table 5.
Performance results for predicting the E (GPa) in the testing phase.
Fig 14.
Measured and predicted UCS (MPa) for the rocks by the XGBoost, MLP, SVR, RF and stacking ensemble in the a) training phase and b) testing phase.
Fig 15.
Measured and predicted E (GPa) for the rocks by the XGBoost, MLP, SVR, RF and stacking ensemble in the a) training phase and b) testing phase.
Fig 16.
Scatter plots of the output of UCS for the a) XGBoost, b) MLP, c) SVR, d) RF, and e) Stacking Ensemble.
Fig 17.
Scatter plots of the output of E for the a) XGBoost, b) MLP, c) SVR, d) RF, and e) Stacking Ensemble.
Table 6.
Training time of various models.
Fig 18.
RSE of input variables on the a) UCS (MPa), and b) E (GPa).