A novel hybrid gated multilayer perceptron and random forest model to predict the compressive strength of heated rocks

Accurate prediction of the compressive strength of rocks after high-temperature treatment is crucial for ensuring the safety and performance of geotechnical structures, including geothermal energy systems, nuclear waste disposal, and deep mining operations. In this study, a new hybrid gated multilayer perceptron–random forest (GatedMLP-RF) model is proposed to predict compressive strength based on seven input parameters: density, porosity, elastic modulus, crack damage stress, temperature, sample diameter, and height. The GatedMLP component performs adaptive feature reweighting, capturing nonlinear interactions within the input parameters, while the RF model improves robustness and interpretability. Model performance was evaluated using five metrics: mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE), variance accounted for (VAF), and coefficient of correlation (R 2 ). The proposed GatedMLP-RF hybrid model achieved an R² of 0.985, an RMSE of 17.71, an MAE of 11.72, an MAPE of 16.80, and a VAF of 98%, outperforming standalone RF, MLP, and GatedMLP models. Additionally, SHAP analysis revealed that crack damage stress and porosity are the most significant influential input features in predicting compressive strength. These outcomes suggest that the proposed hybrid model provides a more accurate and interpretable framework for predicting the compressive strength of heated rocks compared to traditional machine learning methods.