Abstract: At present, there is no fixed way to predict the sensitive zones of subsidence disaster development in underground coal mining areas, and the prediction result of sensitive zones has a great uncertainty. Herein, the subsidence disaster in Xishan area of Taiyuan City, Shanxi Province was taken as the research object. Totally 4 types of kernel SVM based prediction model for sensitivity zoning of subsidence disaster were constructed with the methods of GIS spatial analysis, statistical analysis and Support Vector Machine (SVM) in combination, taking the subsidence disaster data checked and recorded in 2012 and 2014 as the modeling and verification data respectively, as well as the elevation, slope gradient, slope aspect, topographic relief, surface curvature, stratigraphic strata and geological structure as the sensitivity assessment factors. Meanwhile, analysis was performed on the weight of assessment factors, the model optimization, the prediction results of sensitivity zoning, the prediction accuracy, and the applicability of models respectively. The results show that the polynomial kernel-SVM model (PL-SVM) has relatively high training accuracy (with the area under the receiver characteristic curve of AUC=0.854) and validation accuracy (AUC=0.755), as well as good prediction capability. Thus, it has the best performance among the 4 types of models, and the sensitivity zoning is reasonable, with more points of subsidence disaster distributed in a small area of the very-high and high sensitive zones, while few points of subsidence disaster distributed in a large area of the low sensitive zones. As predicted by the PL-SVM model, the area proportion of very-high, high, moderate and low sensitive zones of subsidence disaster in Taiyuan Xishan area is 20.19%, 17.43%, 21.18% and 41.20%, respectively. Besides, the frequency ratio and the sensitivity grade are in good positive correlation, showing a linear functional relation. The sensitivity assessment result based on PL-SVM model is reliable and has good applicability, which has reference significance to the study on the development characteristics of subsidence disasters in underground coal mining areas and the prediction of key areas in disaster survey.