As mineral resource development moves deeper into the earth, mine dynamic disasters, represented by rockbursts, occur frequently. Due to the combined effects of in situ rock stress state and geostress conditions, it is difficult to obtain reliable prediction results using traditional empirical criteria or single-index prediction methods. To address these issues, this paper constructs a rockburst sample database based on the existing literature, including maximum tangential stress, uniaxial compressive strength, uniaxial tensile strength, elastic energy index, stress coefficient, and brittleness coefficient. Secondly, six typical machine learning algorithms are selected for rockburst level classification research. Then, to address the problem of imbalanced sample distribution, the SMOTE oversampling method is introduced to balance the data, and Bayesian optimization and cross-validation are combined to optimize the model hyperparameters. The results show that optimized XGBoost models exhibit high accuracy and stability in rockburst level discrimination, and accuracy reached 0.7664, recall was 0.7664, macro-P was 0.7664, and macro-F1 was 0.7662. Furthermore, taking the BO-XGBoost model as an example, the SHAP method is introduced to analyze the interpretability of the model’s prediction results. The results show that the elastic energy index and stress-related indices are the main controlling factors affecting the intensity of rockburst; they accounted for 25.75% and 22.68%, respectively. Based on the above research results, this paper further explores the ideas for rockburst safety management and prevention from the aspects of energy control, stress regulation, and optimization of rock mass structural characteristics, providing theoretical basis and technical support for the scientific formulation of rockburst risk identification, level prediction, and safety prevention and control measures in deep underground engineering.
Predicting underground mining rockbursts remains challenging because geotechnical data sets are often small, sparse, incomplete, and noisy, resulting from inconsistent testing standards, high acquisition costs, measurement errors, and geological variability. Consequently, traditional machine learning (ML) models frequently exhibit poor predictive accuracy and reliability. To address these issues, this study introduces the Tabular Prior-Data Fitted Network (TabPFN), a novel pretrained transformer that leverages approximate Bayesian inference to generalize effectively from small, noisy data sets. Moreover, this model can natively handle missing values, mitigate overfitting, reduce computational time, and simplify practical deployment, all without requiring fine-tuning. The model was evaluated on 150 rockburst cases using five key parameters to predict four rockburst intensity classes. To quantify its advantages over traditional ML methods, TabPFN was systematically benchmarked against support vector machine, multilayer perceptron, decision tree, random forest, and extreme gradient boosting. The results demonstrated TabPFN’s superior predictive accuracy on noisy and imbalanced data sets, and this advantage persisted in limited-data scenarios. It also performed robustly on incomplete data sets, with further gains when paired with imputation methods like
k
-nearest neighbors. Furthermore, the model provided strong interpretability, identifying the elastic energy index as the most critical predictive factor, followed by tangential stress and buried depth. TabPFN’s practical utility was further confirmed through successful application to real-world engineering projects (93% accuracy). Ultimately, this study establishes TabPFN as a powerful and reliable tool for rockburst prediction and highlights the broader potential of this pretrained model to effectively address the persistent challenge of scarce, poor-quality data in engineering geology.
Ngoc Hai Dong, Kexin Yin, Qing-Wei Zhai· Journal of computing in civi...· 0 citations
Rockburst risk in seismically active mines poses a significant threat to underground safety. This study aims to improve the prediction of rockburst-induced damage by addressing the challenge of class imbalance, which is commonly encountered in rockburst datasets. Various data-balancing techniques, including the Synthetic Minority Oversampling Technique (SMOTE) and its variants, namely SMOTE-Tomek, KM-SMOTE, SMOTE-ENN, SVM-SMOTE, Borderline-SMOTE, and Adaptive Synthetic (ADASYN) sampling, were applied to a dataset containing rockburst damage scales and selected influencing parameters. The data were collected from Canadian and Australian underground mining operations affected by mining-induced seismicity. Three machine learning classifiers, namely Random Forest (RF), CatBoost (CB), and Gradient Boosting (GB), were trained and evaluated using accuracy, recall, precision, and F1-score metrics. The five best-performing models were SMOTE-ENN-GB, SMOTE-ENN-CB, SMOTE-ENN-RF, KM-SMOTE-CB, and Borderline-SMOTE-CB, achieving testing accuracies ranging from 70% to 88%. SHAP analysis further revealed that stress conditions and peak particle velocity (PPV) are the dominant factors controlling rockburst severity, while geological factors and support conditions act as secondary contributing factors. Compared with previous studies using the same dataset, the proposed approach achieved substantial improvements in predictive performance, particularly for the minority and severe rockburst classes. It is concluded that SMOTE-based balancing techniques, when combined with ensemble learning algorithms, can significantly improve rockburst damage prediction and contribute to safer and more effective risk management in deep, seismically active mining environments.
Kairat Sarsembayev, A. Adoko, Rashid Afshar et al.· Applied Sciences· 0 citations
For safe and cost effective design of geotechnical and rock engineering schemes, accurate Uniaxial Compressive Strength (UCS) prediction is crucial, especially in carbonate rock formations with considerable heterogeneity. For enhancing the precision and resilience of UCS prediction, this study proposes a hybrid intelligent modelling framework that combines Aquila Optimization (AO) with a Deep Feedforward Neural Network (DFNN). Comprehensive data collected from laboratory-tested carbonate rock samples are preprocessed, which includes handling missing values and data cleaning. To comprehend the behavior and correlations of input variables, exploratory data visualization and feature distribution analysis are carried out. The regression model is a deep feedforward neural network and important hyperparameters like number of hidden layers, neurons, learning rate and activation functions are optimally tuned using Aquila Optimisation Algorithm (AOA). Multiple regression metrics are used to quantitatively assess model performance from python software. The findings show that proposed framework attains minimum RMSE and MSE value of 0.0425,0.0018 is useful for rock mechanics and mining applications and gives a dependable, data-driven tool for UCS prediction in carbonate rocks.
J. M. Durga, Vinod Kumar Yarlanki, Dasari Appaji et al.· International Conference on...· 0 citations
The significance of rock brittleness is well‐recognized in the fields of geotechnical engineering and energy exploration. To enhance the predictive precision of rock brittleness, this paper proposes a Stacking integrated algorithm. This algorithm synergistically combines various meta models and foundational models, utilizing a suite of nine algorithm modes: Gaussian Process Regression, Support Vector Machine, Backpropagation Neural network, Extreme Learning Machine network, Decision Tree, Random Forest, Extreme Gradient Boosting, Lasso Regression, and Ridge Regression. Furthermore, Tuna and Bayesian optimization algorithms are utilized to refine the model's performance. Additionally, a new diversity index, k, based on the ratio of correlation coefficients, has been introduced to facilitate the optimal selection of base models for the Stacking integrated algorithm. The predictive accuracy of the Stacking integrated model, as determined by the proposed diversity index k, surpasses that of the finest individual base model and outperforms other sets of five integrated base models with an equivalent number of components. This underscores the efficacy of the diversity index k in guiding the selection of appropriate base models for the stacking process. The most effective model for predicting rock brittleness incorporates the Extreme Learning Machine network, Decision Tree, Random Forest and Extreme Gradient Boosting. This ensemble model demonstrates superior accuracy over the single best model, the Decision Tree, by reducing the average brittleness prediction error rate by 0.3745 and elevating the determination coefficient (
R
2
) value from 0.9118 to 0.9629.When compared with the Particle Swarm Optimization model, this composite model achieves an increase of 0.1498 in
R
2
.
Jing Jia, Diquan Li, Ziyi Zhang et al.· International journal for nu...· 0 citations
Rockburst is a dynamic geological hazard characterized by the sudden release of elastic strain energy in underground excavations, posing severe threats to personnel safety. Machine learning (machine learning)-based rockburst prediction using microseismic monitoring faces a critical challenge: class imbalance, where strong rockburst events—the most destructive—are severely underrepresented, leading to poor minority-class predictions. To address this, this study proposes a two-step intelligent framework integrating generative artificial intelligence with deep ensemble learning. First, a conditional tabular generative adversarial network synthesizes high-fidelity samples for minority classes, transforming the imbalanced dataset into a balanced one. Second, a deep random forest is constructed on the augmented dataset to learn the nonlinear mapping between six precursory microseismic parameters—cumulative number of events, cumulative energy, cumulative apparent volume, and their rates of change—and rockburst intensity levels. The whale optimization algorithm enables automatic hyperparameter optimization. Validated on 93 rockburst cases from Jinping II Hydropower Station, the CTGAN-DRF framework increases accuracy from 65% to 90% and Kappa coefficient from 0.53 to 0.87 compared with baseline models.
F
1
-scores for slight, moderate, and strong rockburst improve by 0.29, 0.28, and 0.39, respectively. Comparative analysis confirms the superiority of the deep ensemble architecture over shallow alternatives. SHAP-based interpretability analysis reveals that cumulative parameters dominate lower-intensity predictions, while event count rate becomes critical for strong rockburst. The proposed framework provides a reliable and interpretable solution for rockburst prediction under class imbalance, with promising applications in underground engineering safety.
Hai-Rui Liu, Pei-Tao Li, Yong-Qiang Yu et al.· Frontiers in Earth Science· 0 citations
Accurate and rapid assessment of seismic intensity is crucial for postearthquake emergency response. This is especially true during the initial “black‐box” period, when real‐time data are scarce. To address this challenge, we developed a machine learning‐based (ML) framework for predicting kilometer‐grid‐scale seismic intensity distribution. We selected 50 destructive earthquakes that struck western China after 2003 as case studies and then constructed a high‐dimensional covariate dataset for them. The dataset integrates 39 predictors tied to mainshock intensity, covering seismic parameters, socioenvironmental indicators, and natural geographic attributes. Feature importance ranking was employed to select the top predictive covariates, which were then combined with digitized observed intensity values to train four ML models. The optimized Random Forest model achieved the best prediction performance. Its overall accuracy ranged from 85.3% to 98.7% on independent validation earthquake cases. The corresponding RMSE was 0.21–0.39, and the
R
2
reached 0.88–0.97. Evaluations based on independent earthquake cases indicate that the predictions can reasonably capture the location, extent, and severity of heavily damaged zones. This approach offers a promising, data‐adaptive supplementary tool for rapid postearthquake damage assessment, particularly in regions with limited seismic monitoring infrastructure.
Ma Yuan, Jian-Ming Yu, Zhao Hui et al.· Engineering Reports· 0 citations