CatBoost-SHAP analysis decodes spatial synergies driving prehistoric architectural elongation on the Loess Plateau

Abstract Traditional architectural morphology often overlooks the complex spatial synergies driving evolution. This study proposed a transferable framework integrating systematic architectural plan feature encoding, micro-scale morphological interpretation, and synergistic combination prototype mapping via the CatBoost-SHAP algorithm. Applied to 103 prehistoric dwellings (5000–2000 BC) from the Loess Plateau, China, the analysis identified specific synergy—notably the combination of doorway form and doorway orientation—as the primary catalyst for the model to predict elongation. By translating high-dimensional feature interactions into visualized spatial prototypes, this research offers a robust and replicable pathway for decoding the latent spatial logic of built heritage.