Abstract: To assess the risk of landslide geological hazards induced by slope-cutting house construction, this study takes case in the completely weathered granite slope area of Yanling County, Hunan Province as data samples. Intelligent analysis of the vulnerability of slope-cutting construction was conducted by integrating unsupervised learning algorithms with numerical simulation based on the smoothed particle hydrodynamics-finite element method (SPH-FEM). Vulnerability classification of slope-cutting construction was implemented via PCA dimensionality reduction and KMeans clustering, leading to the identification of key vulnerability factors. Furthermore, the SPH-FEM numerical model was employed to reveal the motion-accumulation relationships and energy transfer laws between landslide hazards and residential structures under varying levels of these key factors. The results indicate that slope-cutting height and the ratio of retaining wall height to slope toe distance (H/D_s) are the critical vulnerability factors. Plotting the cumulative percentage curve of pca1 for clustering-derived moderately vulnerable samples enables further subdivision of moderately vulnerable slope-cutting construction cases to support hierarchical risk control. When the slope toe distance (D_s) is≤0.5 m, a substantial amount of kinetic energy from the sliding mass is transferred to the retaining wall, resulting in severe damage. As D_s increases, the energy absorption capacity of the wall decreases, and the hazard intensity is significantly mitigated. When D_s≥6 m, the wall only undergoes minor deformation, which can be regarded as the safe slope toe distance under the studied scenarios.