Identifying drivers of community-scale urban renewal with an ante-hoc interpretable hybrid graph model

The driving mechanisms of urban renewal are complex and spatially heterogeneous, particularly at the community scale. However, existing studies often fail to capture complex neighborhood dependencies and lack the built-in interpretability to decode decision-making logic, representing a critical research gap. This study develops an interpretable framework based on an Graph Convolutional Network (GCN) to identify the drivers of community-scale urban renewal. On one hand, the GCN explicitly models the city as a spatial graph, enabling the capture and aggregation of non-local and irregular neighborhood information, effectively overcoming the limitations of traditional models in handling complex urban spatial dependencies. On the other hand, unlike the common "predict-then-explain" approach, the framework integrates a priori Attention through a Multi-Layer Perceptron module, providing endogenous interpretability. The spatial context features and attention-weighted features are ultimately integrated, providing comprehensive and profound information for prediction. Using Changsha, China as a case, the model achieves an 84 % accuracy in Land use transition prediction, outperforming the classical XGBoost model by 10 %. Its ante-hoc interpretability reveals that the primary driving forces of urban regeneration are Land use and the Environment (2018: 0.358 and 0.275; 2024: 0.397 and 0.247).The attention weight map decodes the spatial heterogeneity of these drivers at the community scale: the economy is decisive in central business districts, while the environment dominates in newly developed areas. This framework provides a powerful and transparent tool for understanding complex urban systems, enabling planners to shift from "experience-driven" to "data-driven" decision-making. • Ante-hoc interpretable GCN predicts urban renewal with 10 % accuracy over classical XGBoost. • GCN model identifies the implicit path of main functional transformation. • Landuse (weight 0.358) and Environment (weight 0.275) are dominant drivers of urban renewal in Changsha. • 300m community scale reveals significant spatial heterogeneity: distinct renewal logic across zones.