An Integrated Spatial Renewal Strategy for Historic Districts Using Machine Learning and Perceptual Evaluation: A Case Study of Subtropical Climate

The renewal of historic districts requires a balance between heritage conservation, spatial quality improvement, and contemporary urban use. Yet the links between perceptual experience and measurable spatial characteristics remain insufficiently clarified. Using the Yantai Mountain Historic District in Fuzhou, China, as an empirical case, this research develops an integrated evaluation framework that combines perceptual assessment, street-view image analysis, and interpretable machine learning. Street view images, field investigations, and questionnaires were used to construct a subjective-objective evaluation system. A total of 1408 sampling points were established, and 1272 valid street view images were collected. Perceptual scores were provided by 15 local residents and 15 professionals. Objective indicators, including enclosure, greenery, openness, motorization, and non-motorization, were extracted through semantic segmentation. Multiple linear regression and random forest models were then used to examine the relationships between spatial indicators and perceptual experience. The findings reveal nonlinear associations and multi-factor interactions between objective spatial characteristics and subjective perception. Safety and cultural perception scores are higher when the enclosure ranges from 0.30 to 0.60, and the non-motorization ranges from 0.10 to 0.30. Comfort and visual pleasantness scores are higher when non-motorization ranges from 0.10 to 0.20 and motorization from 0.05 to 0.15. SHAP analysis further indicates that openness and motorization primarily influence perceived safety, whereas non-motorization and enclosure contribute more strongly to cultural perception, comfort, and visual pleasantness. The proposed framework provides a quantitative and interpretable basis for refined spatial renewal in historic districts.