In Japan, numerous landslides are triggered by heavy rainfall. In 2024, 1,433 landslides were recorded, resulting in 56 fatalities. To protect lives and property from such slope hazards, it is crucial to establish a framework that can accurately and rapidly predict landslide occurrences over wide areas. Previous studies have primarily relied on statistical approaches using slope-related information or empirical prediction models based on the relationship between rainfall and past landslide records. However, the integration of slope-scale mechanical numerical analyses remains limited. This study proposes a framework for rapid slope hazard assessment that accounts for rainfall-induced infiltration by integrating numerical analysis, observed data, and a machine learning approach, and examines the validity of the developed model. For the numerical analysis, a three-dimensional limit equilibrium method is employed to evaluate slope stability by calculating the factor of safety. In addition, an infiltration analysis is coupled with the stability analysis to compute the factor of safety under rainfall conditions. To enable efficient large-scale evaluation, a surrogate model is constructed using spatial mode decomposition. A database of landslide hazard distributions is generated through pre-analyses under various rainfall scenarios. By applying mode decomposition to these results, the dominant spatial characteristics of the landslide hazard distributions are extracted and combined with observational data to construct a surrogate model for rapid identification of potentially hazardous slopes. The proposed framework is applied to a specific city in Japan, and the results obtained from the surrogate model are compared with those from the corresponding numerical analysis results. The comparison indicates that the proposed method can reasonably reproduce the numerical analysis results while achieving a reduction in computational cost.