This paper is the result of geological survey engineering.

Objective To overcome the limitations of traditional global modeling in effectively revealing the spatial differentiation patterns of regional geological hazards, this study proposes a geological hazard susceptibility assessment method that integrates spatial heterogeneity zoning with feature factor optimization.

Methods Chaoyang City, Liaoning Province, was selected as the study area. A 1-4 level spatial zoning system was constructed by combining spatial clustering results of hazard points with Thiessen polygons. The Fick’s Law _ K Nearest Neighbor (FLA_KNN) algorithm was then applied to select optimal feature factor subsets for each sub-region. Furthermore, a Particle Swarm Optimization _ Random Forest (PSO_RF) model and a Sparrow Search Algorithm _ Support Vector Machine (SSA_SVM) model were employed for susceptibility zoning prediction and comparative analysis.

Results The results indicate that: (1) Spatial partitioning improved model accuracy by approximately 7%, and prediction performance exhibited a “first increase and then decrease” trend with increasing partition levels, with the 3-level partition achieving the best performance (AUC = 0.8897).(2) The average AUC of the Partition Feature Selection (PFS) model was 6.53% and 2.26% higher than those of the Global Feature Selection (GFS) model and the Partition Uniform Feature (PUF) model, respectively, demonstrating the synergistic advantages of spatial adaptation and feature factor optimization.(3) The overall performance of the PSO_RF model surpassed that of the SSA_SVM model, and the prediction discrepancy between the two models was smallest under the 3-level partitioning scheme. (4) Partition II was identified as a high-risk hazard zone, primarily influenced by rainfall, lithology, topographic wetness index (TWI), and human engineering activities.

Conclusions The proposed method effectively enhances the accuracy of geological hazard susceptibility assessment and provides scientific support for regional hazard prevention, mitigation, and spatial planning.