Prediction Model for Indoor Rock Compression Failure Time Based on Ensemble Learning and Optimization Algorithms

Predicting the time of rock failure is a fundamental predictive model in the prediction of mine slope instability as well as in rock mechanics research.Predicting rock failure time is a fundamental predictive model in engineering project safety assessment, construction efficiency optimization, environmental risk assessment, and rock mechanics research. Traditional methods struggle to effectively describe the entire process of rock failure, while machine learning provides a new solution. This paper establishes 12 prediction models based on ensemble learning and optimization algorithms, sets the strain of the test set, and then predicts rock stress and failure time. First, a dataset was established through rock mechanics experiments, and suitable algorithms were selected to build models. The test set strain increments were configured at 0.008‰, 0.01‰, and 0.012‰. Five-fold cross-validation was used to optimize hyperparameters, significantly improving the model's generalization ability, robustness, and stability. These models were used to predict rock stress, and the peak stress and its corresponding strain were used as input parameters to further predict rock failure time. Through evaluation metrics, it was found that the Particle Swarm Optimization algorithm combined with Extreme Gradient Boosting (PSO-XGBoost) performed best in predicting rock stress and failure time. When the strain increment was 0.01‰, the PSO-XGBoost model achieved R²=0.904, MAE=4.315, and RMSE=5.435 for rock stress prediction; for rock failure time prediction, it achieved R²=0.811, MAPE=7.842%, and MAE=30.343, demonstrating the model's excellent performance. Finally, SHapley Additive exPlanations(SHAP) analysis showed strain and stress significantly impact the model, with strain positively predicting failure time, aligning with traditional rock failure models, validating reliability. This study aids rock engineering disaster warning systems.