Abstract: To improve the current situation where ground subsidence susceptibility assessment mainly relies on knowledge-driven models, this study explores the feasibility of incorporating data-driven models into the evaluation of urban ground subsidence. The study focused on a typical area in Hangzhou characterized by fill and silty soil. The selection of ground collapse indicators was conducted, followed by a correlation test. Seven evaluation factors, including drainage pipeline density, social activity density, depth of underground confined water level, thickness of surface fill layer, distance from hidden rivers and beaches, depth of the saturated sand top plate, and thickness of the soft soil layer, were selected for assessing the susceptibility to ground subsidence in the study area. By comparing the Random Forest (RF) model, RF-I integrated model, and RF-BP neural network integrated model, it was found that the integrated model had higher accuracy in assessing the susceptibility of ground collapses subsidence in this study area compared to single models. Ultimately, the RF-BP neural network integrated model, which showed the best performance, was chosen for susceptibility assessment. The assessment results indicated a high correlation between the susceptibility zones and areas prone to ground subsidence, indicating good prediction performance and proving the potential application of data-driven models in evaluating the susceptibility of urban ground collapses.