关键词: 采空区, RF算法, RFE算法, 逻辑回归理论, K-means聚类

Abstract: The study on risk classification of goaf has important significance in mine disaster prevention and risk management.In order to overcome the problems of numerous indexes and complex calculation in traditional methods，a rapid grade evaluation model of goaf is proposed.Based on 110 goaf samples，combines random forest （RF） with recursive feature elimination （RFE） algorithm to select indicators that contribute most information in classification，which overcomes the shortcomings in traditional methods whose indicators which are numerous and difficult to obtain，realizes the dimension reduction of the evaluation index system of goaf.Based on logistic regression theory，the probability model of goaf risk is obtained，and four clustering centers of goaf risk probability are obtained by fast clustering algorithm，coupled with two algorithms，a fast grade evaluation model of goaf is constructed to overcome the shortcomings of complex calculation and poor universality in traditional methods.In order to verify the validity of the evaluation model，its accuracy was verified and analyzed based on confusion matrix.The study results show that:①RQD value，pillar size and layout，rock mass structure，goaf height，geological structure，engineering layout，groundwater are these indicators that contribute most information in goaf risk classification； ②the classification accuracy rate of the fast classification model constructed in this paper reaches 77.4%， the error rate of the first category is as low as 6.25%，and the accuracy rate of predicting dangerous goaf reaches 93.75%， the model can provide effective information for goaf management in actual production.

Key words: Goaf, RF algorithm, RFE algorithm, Logistic regression theory, K-means clustering