Risk-informed Conservation of Mountain Railway Heritage Using GIS and Explainable Machine Learning: Landslide Susceptibility and Exposure along the Late Qing Sichuan-Hankou Railway, China

This dataset provides an integrated spatial and analytical record for the corridor-scale landslide-susceptibility and heritage-exposure assessment of 90 Late Qing Sichuan-Hankou Railway heritage sites in Huanghua, Shuiyuesi, Wuduhe, and Xiakou townships, Yichang, China. The dataset contains 145 landslide-positive samples, 145 non-landslide-negative samples, 28 candidate conditioning factors and their screening results, a 35-feature modelling matrix, repeated five-fold cross-validation outputs for five machine-learning models, Random Forest probability and five-class susceptibility rasters, SHAP values and factor-importance tables, and exposure classifications for all 90 heritage sites. Random Forest achieved an out-of-fold AUC of 0.8101 and a PR-AUC of 0.7779. The exposure assessment identifies 76 sites, accounting for 84.44% of the heritage inventory, within High or Very High susceptibility zones. The archive provides analysis-ready CSV tables, 30 m GeoTIFF rasters in EPSG:4526, GeoPackage spatial databases, a bilingual site-identifier and coordinate crosswalk, and a self-contained offline WebGIS. It also includes variable definitions, model configurations, figure- and table-level source mappings, a complete file manifest, and Python scripts for data export, Random Forest validation and fitting, and summary visualization. Derived factor values and full provenance information are included, while third-party source rasters remain with the providers identified in the variable dictionary. The dataset supports reproducibility, spatial verification, conservation-priority assessment, digital heritage management, and comparative GIS-based research on mountainous linear transport heritage. Data, documentation, derived figures, and the WebGIS are distributed under the Creative Commons Attribution 4.0 International License, while the Python code is distributed under the MIT License.