Abstract: The occurrence conditions of coal seams in western Huanglong Jurassic coalfield are generally thick, of which the average thickness of coal seams in Binchang Mining Area is greater than 5 m, and the thickest coal seam can reach 14 m, and the fully mechanized caving technology is often adopted, resulting in large thickness and unclear development law in the water-conducting fracture zone of coal seam roof, and high water inflow in the mine, which seriously affects the safety production in the mining area. In order to study the development height of seam roof water-conducting fracture zone caused by disturbed overlying rock mining in Binchang Coal Mine, seven influencing factors such as the thickness of coal seam, seam buried depth, roof overlying rock lithology, roof structure characteristics, mining speed, the length of working face and mining technology were first selected. Firstly, the weight of the above influencing factors is calculated by AHP, and it is found that the weight of the two influencing factors, the thickness of coal seam and the length of working face, is relatively large. The collected data is interpolated by Matlab to make the data distribution smoother. Back propagation neural network, genetic algorithm and particle swarm optimization were used to optimize BP neural network and random forest algorithm to carry out regression fitting for the interpolated data. It is found that the four methods have better fitting effect on the original data, and random forest RF has higher fitting accuracy than the other models. The root mean square errors (RMSE) of the training set and the test set are 0.03741 and 0.05516, and the determination coefficient R² is 0.98737 and 0.95789, respectively. The research results can provide some references for predicting the development height of the water-conducting fracture zone in Binchang Coal Mine.