[Objective] A comprehensive risk assessment system for collapses and landslides in the karst regions of the Guangxi Zhuang Autonomous Region was established, in order to offerscientific support for early warning, disaster prevention, and mitigation in the area. [Methods] Beiliu City was selected as the study area, and a database of collapses and landslides was constructed. Slope units were used as the basis for the evaluation, with multisource data systematically collected and analyzed. Key evaluation indicators, including groundwater type and runoff intensity index, were identified, and a fully connected neural network model was employed to assess the susceptibility to collapses and landslides. Given the region’s vulnerability to rainfall and karst erosion, the soil erosion modulus was incorporated into the hazard assessment. Finally, a risk evaluation model for collapses and landslides in Beiliu City was developed by integrating vulnerability assessments of the exposed elements. [Results] The findings revealed that high- and very high-risk zones covered 252.22 km², accounting for 10.27% of Beiliu City’s total area. These zones are primarily located in Longsheng Town, Xinfeng Town, Pingzheng Town, and Liujing Town, which are characterized by eroded and denuded hills and tectonic erosion of low mountains. Factors such as loose geotechnical body, high soil erosion modulus, dense population, and concentrated buildings significantly heightened the collapse and landslide risks, resulting in a high-risk classification. [Conclusion] Validation through ROC curves and field investigations showed an accuracy of 0.966 4 for susceptibility evaluation and 89.3% for risk assessment in Beiliu City. These results demonstrate the high precision and practical applicability of the constructed model, which aligns closely with real-world scenarios.