Abstract: Landslide has always been an important issue in engineering geology; most studies focus on landslide disasters in mountainous areas, few studies focus on landslide disasters in the loess plateau area． Loess plateau, an important geographic unit in northern China, has special geological structures, under certain climatic conditions often leads to landslides, mudslides and other disasters, causing great disasters and property losses． In this paper, landslide disaster in loess plateau geomorphology area is studied, to establish a comprehensive quantitative evaluation system for landslide susceptibility by comprehensively considering factors of topography and geomorphology, hydrology, soil and other factors． LightGBM is used to simulate landslide susceptibility of loess plateau． Five indicators are selected to evaluate prediction performance and robustness． SHAP algorithm is used to analyze influence of triggering factors on the simulation results, with mechanism of landslide susceptibility in loess plateau geomorphologic area revealed． LightGBM model could simulate and predict susceptibility of landslides in loess plateau rather well (AUC = 0.844)． SHAP algorithm identifies key driving factors: distance from road, annual average rainfall, degree of topographic undulation, density of population, depth of ground surface cutting, length of slope. Landslide susceptibility in the study area has been mapped． A risk map was drawn. This work provides solid theoretical basis for regional landslide management and landslide disaster prevention and mitigation．