Intelligent Optimization of Public Space Allocation in Urban Renewal Using Ensemble Learning

The deposit contains four site-level datasets covering 3,500 urban renewal parcels in China (2018–2025) — municipal renewal project records, Sentinel-2 and VIIRS remote-sensing composites, a site visitation panel, and resident survey responses aggregated to site level — together with the complete analysis pipeline: leakage ablation, AHP weight sensitivity, spatially blocked cross-validation with GWR and spatial lag/error benchmarks, and external validation by region and climate-zone hold-out. All random seeds are fixed, so every table and figure the article reports from these experiments regenerates from the deposited code; key values to check are a leakage-free composite R² of 0.875, region hold-out R² of 0.860–0.884, and climate-zone hold-out R² of 0.837–0.858. The resident survey is released only as site-level aggregates so that no respondent can be identified. Data are licensed CC BY 4.0 and code under the MIT Licence.