Ripples Create Waves: Integrating Interpretable Machine Learning and Econometrics to Enhance Street-level Vitality through Neighborhood Micro-Regeneration

Keywords: Micro-regeneration, Street-level vitality (SLV), Spatial configurational, Interpretable machine learning, Urban sustainability, Pedestrian movement
Abstract Type: Paper Abstract

Abstract

Shifting from large-scale redevelopment to small-scale, tenure-secure regeneration is crucial for promoting sustainability and social equity in neighborhoods. Place-based strategies can invigorate local economies and create living-wage jobs for low-income populations, serving as new growth drivers instead of relying on land exchange value. While predicting street-level vitality (SLV) through configurational relationships is well-established in spatial syntax models, micro-regeneration scenarios—which preserve urban fabric and street networks—often lead to underfitting due to their focus on place-based strategy in sporadic locations. Yet the impacts of external attractors like culture, community facilities and public spaces and their combinatory effects on SLV still remains not clear-cut. This study develops a framework using interpretable machine learning and econometrics to predict and explain the spatial variance of SLV in Nantou Ancient Town (NAT), a micro-regeneration pilot in China, before and after intervention: (1) micro-regeneration efforts, though small-scale and not altering the physical grid, can significantly enhance SLV, as demonstrated by Difference-in-Differences comparisons of commercial establishments between NAT and a control group (2) Incorporating the attractor effect from micro-regeneration substantially improves predictive power, increasing the R² by 0.31 to achieve 0.83 using the XGBoost model. (3) Combining metric and configurational properties in interpretable machine learning and econometric models enhances the ability to predict and explain SLV before and after micro-regeneration. (4) Among the configurational properties analyzed, angular integration (closeness centrality) exhibits the strongest explanatory power, directly influencing SLV and suppressing other configurational variables. These insights offer predictive methods and guidance for urban planners and policymakers to implement micro-regeneration。