Abstract: Currently, there are numerous methods for evaluating rock brittleness index, mainly based on mineral composition or rock mechanical properties evaluation, but most of these evaluation indicators are costly to obtain and time-consuming. By utilizing machine learning techniques, a rock brittleness index prediction method based on the Stacking ensemble learning concept is proposed. This method involves parallel training of Gradient Boosting Decision Tree model (GBDT), Random Forest model (RF), Naive Decision Tree model (DT),Support Vector Regression model (SVR), and LightGBM model. After hyperparameter tuning using a tree-structured Parzen estimator for each model, the XGBoost model is sequentially used to merge the training results of the base models to achieve rapid parameter optimization and prediction of rock brittleness index. The results indicate that the two-layer Stacking model optimized with the tree-structured Parzen estimator shows significant advantages compared to the predictions of the base models. The explained variance score (EVS) reaches up to 0.97, and the coefficient of determination (R²) reaches a maximum of 0.967. In the same dataset performance comparison, this model exhibits the lowest Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), implying that this model can effectively capture the variation patterns of rock brittleness index under the supervised learning framework. This verifies its practical value in predicting rock brittleness index.

Key words: Rock brittleness index, Stacking model, Ensemble learning, Tree-structured Parzen Estimator