The impact of spatial heterogeneity zoning and feature factor optimization on geological hazard susceptibility assessment: a case study of Qingyuan County, Liaoning Province

The spatial distribution of geological hazards exhibits significant heterogeneity and scale dependence. Therefore, a key challenge in susceptibility assessment is to effectively characterize spatial heterogeneity and accurately identify the main controlling factors. To address this, this study introduces a comprehensive framework. Qingyuan County, Liaoning Province, was selected as the study area. Hazard point clustering and Thiessen polygons were used to construct a six-level zoning system. The Fick’s law _ K Nearest Neighbor (FLA_KNN) algorithm was applied to screen optimal feature subsets for each zone, and susceptibility predictions were generated using a Particle Swarm Optimization _ Random Forest (PSO_RF) model. Model performance was evaluated by AUC, Accuracy, Precision, Recall, and MSE. Results show that: (1) Spatial heterogeneity zoning improved accuracy by about 5%, with prediction accuracy following a nonlinear trend across scales. The 4-level scheme achieved the best performance (AUC = 0.923); (2) Feature optimization further enhanced performance, stabilizing at eight features, with profile curvature and rainfall being most important; (3) High-risk areas were concentrated along the southwestern fault zone and central valleys, jointly influenced by slope (5°–20°), tectonic activity (fault density 0–0.1), and human activities (distance to roads 0–500 m). This framework provides a novel approach for susceptibility assessment and supports disaster prevention planning.