摘要
Abstract
The stability of goafs is closely related to the safety of mine production and sur-face structures,and is therefore a key concern in mine disaster prevention and control.Traditional stability assessment methods are often limited by expert judgment and the appli-cability of numerical models,making it difficult to effectively address multi-class classifi-cation problems under complex geological conditions.In this study,a goaf stability dataset consisting of 93 samples was established.Seven representative machine learning algorithms,namely Back Propagation Neural Network,Decision Tree,K-Nearest Neigh-bors,Naive Bayes,Random Forest,Support Vector Machine,and eXtreme Gradient Boosting,were employed for modeling and comparative analysis to investigate differences in multi-class classification performance.The results show that Random Forest,K-Nearest Neighbors,and Support Vector Machine outperform the other models across multiple evalu-ation metrics,demonstrating strong robustness and classification capability.On this basis,an ensemble assessment framework integrating majority voting and weighted priority voting was developed to determine the stability levels of 10 test samples.The overall assessment accuracy reached 80%,comparable to that of the best-performing individual model.Al-though the ensemble method does not show a clear numerical advantage over the optimal single model,it can reduce the influence of occasional misclassifications by individual models under small-sample conditions.In addition,it improves the robustness of classifying samples located near class boundaries,thus providing a relatively reliable auxil-iary decision-making tool for the stability assessment of complex goafs.关键词
机器学习/采空区/稳定性预测/评估指标/多模型融合Key words
machine learning/goaf/stability prediction/evaluation index/multi-model fusion分类
矿业与冶金
