A LightGBM-based landslide susceptibility model considering the uncertainty of non-landslide samples

The quality of samples is crucial in constructing a data-driven landslide susceptibility model. This article aims to construct model that takes into account the selection non-landslide samples. First, 21 conditioning factors are selected, including four types topography and landform, geological conditions, environmental human activities. Grid units with 30 m resolution established by combining 942 historical events study area. Second, selected using both traditional method information quantity method. Two models Bayesian optimization-LightGBM accuracy evaluated significance test area under curve (AUC). Finally, SHAP algorithm used analyse internal mechanism model’s decision-making. Based on method, LightGBM identifies very high-high areas for 77.92% total number landslides. Additionally, AUC set training 23.2% 17.1% higher, respectively, compared different sample data, whether or non-landslide, impacts factor rank, accuracy, interal decision-making finding provides valuable data binary classification