<p>Landslides pose a significant threat to persons in structures, open spaces, and vehicles. However, the assessment of risk for persons in buildings (PsIBs) remains primarily challenged. This study introduced a novel framework to evaluate PsIBs risk under various rainfall scenarios. First, potential landslide sources were identified by multiplying temporal and spatial probabilities. Temporal probability was determined using a physics-based model and Monte Carlo simulation, while spatial probability was estimated using a convolutional neural network (CNN) trained on landslide samples selected through conditioned Latin hypercube sampling (CLHS). Second, the risk to buildings and PsIBs were estimated from a semi-quantitative approach. Lastly, an integrated risk index was formulated by combining the risk indices for buildings and persons therein. The framework was validated using data from the 2014 landslide event at Mt. Abusan in Hiroshima, Japan. The results show that the generated sample of the landslide inventory closely matched the real 2014 inventory in terms of slope distribution, soil depth, geology, and profile curvature. The area under the receiver operating characteristic curve (AUC) for temporal, spatial and hazard probability maps is reliable in assessing landslide risk with 68.8%, 82.5%, and 75.0%, respectively. The landslide risk from the 2014 event aligned with the predicted risk for a 100-year return period. These findings suggest that the proposed framework is a reliable tool for assessing and mitigating landslide risk, applicable in regions with or without existing landslide inventories.</p>