Application of stacking regressor to predict rock fracture toughness mode-I

Predicting the fracture toughness of rocks, particularly under Mode-I loading conditions, is essential for various geotechnical and civil engineering applications. Traditional methods for determining rock fracture toughness (RFT) are often labor-intensive, time-consuming, and prone to inaccuracies due to the inherent variability in rock properties. This study investigates the efficacy of using a stacking regressor, an advanced ensemble learning technique, to predict the Mode-I RFT. In the proposed model, the strengths of multiple base regressors were combined. 400 experimental data points were utilized, obtained using the cracked Chevron notched Brazilian disc (CCNBD) test and comprising six input parameters affecting the Mode-I RFT. The dataset was partitioned into training and validation sets, ensuring rigorous model evaluation. The stacking regressor's meta-model was trained on the outputs of the base models, effectively learning to integrate their predictions to yield a more accurate final prediction. The performance of the stacking regressor was assessed through several statistical metrics. The results demonstrated that the stacking regressor significantly outperforms individual base models, achieving higher predictive accuracy and reliability. A sensitivity analysis using the mutual information test (MIT) method revealed that the uniaxial tensile strength (UCS) exerts the most significant influence on the Mode-I RFT, underscoring its importance in predictive modeling. Furthermore, developing a machine learning-based graphical user interface (GUI) enhanced the practical applicability of the proposed model, making it accessible to engineers and researchers without extensive expertise in machine learning.