The research on land use and travel safety is a hot topic of common concern in the fields of geography and transportation. However, currently, research on the impact of land use on traffic accidents is mostly included in the unified framework of the built environment, often measured by the degree of land use mixing or the proportion of land use types, lacking detailed research on land use types. This article is based on pedestrian traffic accident data and POI data from Yuzhong District, Chongqing from 2010 to 2021. Using the Extreme Gradient Enhancement Decision Tree (XGBoost) model, combined with road conditions and road environment, the relationship between land use pattern in the accident buffer zone and the severity of pedestrian traffic accidents is explored. The results indicate that the land use pattern plays an important role in the severity of pedestrian walking accidents, with the most influential factors being hospitals, residential areas, and educational land. Through SHAP attribution analysis, it was found that the presence of curves and ramps can increase the severity of traffic accidents, while the presence of hospitals, residential areas, and educational land within 300 meters of the accident buffer zone can reduce the severity of accidents. This study helps to enrich the research on the relationship between land use and pedestrian traffic accidents, and has reference significance for optimizing land use, improving the built environment and reducing the severity of pedestrian traffic accidents.