Recently, 3D seismic technology has become an important method in coalfield exploration, and great progress has been achieved. However, the prediction of water content in coal seam roof using 3D seismic technology is rarely discussed. In this study, a neural network inversion was carried out based on the logging constrained inversion. Seismic attributes, such as wave impedance, were converted to porosity and resistivity, which were closely related to water content. Taking the mining area of Huaibei coalfield as a case study, the water abundance of 10# coal seam roof in the study region was predicted by the neural network inversion of porosity and resistivity. The results show that a water-rich subsided column develops in the north of mining area, which is consistent with the borehole detection results. There is unconformity contact between the 10# coal seam roof and the Quaternary aquifer in the west of mining area, which is predicted as a water-abundant area by the PPN inversion. The neural network inversion can effectively predict the water abundance characteristics of a coal seam roof, thus providing an important guarantee for coalmine safety production.