Abstract: Previous studies have focused on the height of water-conducting fracture zones in single coal seam mining, but there has been little research on predicting the height of water-conducting fracture zones in overburden rock during mining of extremely thick coal seams. This article takes the working face (9-15) of the Luanhuagou Coal Mine in the southern Xinjiang coalfield as the research area, quantitatively evaluates the development characteristics and evolution laws of the overburden rock fracture field under the condition of fully mechanized top-coal caving mining in extremely thick coal seams, and uses machine learning methods to construct a water-conducting fracture zone height prediction model based on particle swarm optimization algorithm support vector regression (PSO-SVR).Research shows that the overall evolution of fractures in the layered fully mechanized top-coal caving mining of a thick coal seam working face generally presents four stages: the rising dimension stage, the decreasing dimension stage, the stable stage, and the fluctuating stage.Among them, the fractal dimension rises rapidly due to the breakage and collapse of the roof overburden affected by mining.However, the fractal dimension of the overlying rock gradually decreases due to compaction.In addition, the correlation coefficient R of the PSO-SVR model is greater than 0.95, and the average absolute error, average deviation, and root mean square error are small. The absolute error between the model prediction value and the measured value is 12.52 m, and the relative error is 4.86%. This indicates that the PSO-SVR model can effectively and accurately predict the height of the water-conducting fracture zone in the mining of thick coal seams.