Abstract: The ground movement and deformation caused by underground coal mining have caused different degrees of damage to the buildings in the affected area. The traditional evaluation method is to evaluate the damage according to the expected horizontal deformation,slope and curvature,other methods,such as fuzzy comprehensive evaluation,cluster analysis and so on,do not take into account the factors of the building itself,and are subjective to some extent,it can not be used as a basis to judge the rationality of current mining and to obtain compensation for damage to buildings. Based on the prediction of building damage grade in traditional mining area,this paper comprehensively analyzes the influencing factors of building damage,the curvature,horizontal deformation,construction time,structure and area which affect the damage of buildings in mining area are selected by consulting literatures and data analysis. Taking the buildings of two villages in a mining area of Hebei province as the research target,the data set of 314 buildings selected from one village and the survey results of the actual damage were used as the training sample data set and carried on the housing survey grading map,training and application of stochastic forest model. Using survey grading to measure the accuracy of evaluation,70 of which were predicted correctly by traditional methods,and 235 of which were predicted correctly by random forest,the accuracy was improved obviously. The model was then used to rank the data sets of 278 test samples from another village,197 of which were consistent with actual damage survey ratings,compared with 117 of which were predicted correctly using traditional methods. The results show that compared with the traditional method,the evaluation precision of this method is obviously improved,and it is more suitable to the actual damage of buildings,it is found that the training sample index is not comprehensive and the grade distribution is uneven,which has some influence on the final prediction accuracy.