Abstract: In order to prevent stope safety accidents, 10 factors including underground stope's geological structure, burial depth, surrounding rock strength and rock mass quality indexes are selected as the prediction indexes of stope stability, twenty-five groups are selected as training samples from the collected data on actual stope stability to build a random forest model of stope stability prediction and eight sets of test data are used to predict stope stability. A prediction result comparison is made between the model, SVM model and ANN model. The results show that the random forest model has the highest accuracy of the stope stability ratings in the examples, and the second is the SVM model, the ANN model is less accurate, and that by using the random forest model, the stope stability can be determined more effectively.

Key words: random forest(RF) model, stope stability, support vector machine(SVM), predictive accuracy, artificial neural network(ANN)