Scholarlink

<p>Coal burst is a significant geohazard that can seriously impact the safety and production of underground coal mines. Accurate prediction of coal burst liability (CBL) is crucial for preventing and mitigating its potentially destructive impacts. For this purpose, this study proposes five white-box models for intelligent prediction of CBL based on four key variables affecting coal burst, including dynamic failure time (DT), elastic energy index (EEI), impact energy index (IE), and uniaxial compressive strength (UCS). The models were developed using five machine learning algorithms, including J48 decision tree, logistic model tree, multivariate adaptive regression splines, tree-augmented naive Bayes, and classification and regression tree. A database containing 206 actual coal burst cases was used as the original supporting database for developing models, and synthetic minority oversampling technique (SMOTE) was applied to address data imbalance and improve the accuracy of CBL prediction. The optimal (most efficient) model was selected based on six performance metrics namely accuracy (ACC), sensitivity (SE), precision (PR), Matthew’s correlation coefficient (MCC), Cohen’s Kappa coefficient (Kappa), and area under the receiver operating characteristic (ROC) curve (AUC). The results indicated that the tree-augmented naive Bayes model demonstrates superior prediction performance with ACC = 0.952, SE = 0.952, PR = 0.952, MCC = 0.929, Kappa = 0.929, and AUC = 0.964. To determine the influence of the input parameters and evaluate the contribution of each variable to the model’s predictions, Shapley Additive exPlanations (SHAP) analysis was used to explain the developed tree-augmented naive Bayes model. The SHAP feature importance quantification results revealed that UCS exerts the most significant influence on the prediction of CBL, followed by the IE and EEI, while the impact of the DT is the least. Finally, the engineering applicability of the tree-augmented naive Bayes model was verified by 16 coal burst cases, achieving a prediction accuracy of 93.75%. The tree-augmented naive Bayes model, by providing an accurate prediction, offers an efficient framework for safety management and coal burst risk mitigation in underground coal mines.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Developing predictive models to assess coal burst phenomena using machine learning algorithms

  • 期刊论文

  • 发表日期：

    2026年2月23日

查看全文
  • Farzad Eslami,
  • Ebrahim Ghasemi,
  • Mohammad Hossein Kadkhodaei

摘要

Coal burst is a significant geohazard that can seriously impact the safety and production of underground coal mines. Accurate prediction of coal burst liability (CBL) is crucial for preventing and mitigating its potentially destructive impacts. For this purpose, this study proposes five white-box models for intelligent prediction of CBL based on four key variables affecting coal burst, including dynamic failure time (DT), elastic energy index (EEI), impact energy index (IE), and uniaxial compressive strength (UCS). The models were developed using five machine learning algorithms, including J48 decision tree, logistic model tree, multivariate adaptive regression splines, tree-augmented naive Bayes, and classification and regression tree. A database containing 206 actual coal burst cases was used as the original supporting database for developing models, and synthetic minority oversampling technique (SMOTE) was applied to address data imbalance and improve the accuracy of CBL prediction. The optimal (most efficient) model was selected based on six performance metrics namely accuracy (ACC), sensitivity (SE), precision (PR), Matthew’s correlation coefficient (MCC), Cohen’s Kappa coefficient (Kappa), and area under the receiver operating characteristic (ROC) curve (AUC). The results indicated that the tree-augmented naive Bayes model demonstrates superior prediction performance with ACC = 0.952, SE = 0.952, PR = 0.952, MCC = 0.929, Kappa = 0.929, and AUC = 0.964. To determine the influence of the input parameters and evaluate the contribution of each variable to the model’s predictions, Shapley Additive exPlanations (SHAP) analysis was used to explain the developed tree-augmented naive Bayes model. The SHAP feature importance quantification results revealed that UCS exerts the most significant influence on the prediction of CBL, followed by the IE and EEI, while the impact of the DT is the least. Finally, the engineering applicability of the tree-augmented naive Bayes model was verified by 16 coal burst cases, achieving a prediction accuracy of 93.75%. The tree-augmented naive Bayes model, by providing an accurate prediction, offers an efficient framework for safety management and coal burst risk mitigation in underground coal mines.