Data-Driven Retrofit Decision-Making: Integrating BIM and XGBoost for Surface-Level Thermal Performance

Improving the energy performance of existing buildings is essential for reducing operational energy use and meeting long-term climate targets. However, assessing retrofit potential at a level of detail suitable for targeted interventions remains challenging, as it often requires extensive manual preparation of geometric and spatial data. This creates a practical barrier to applying data driven retrofit analysis in everyday design and assessment workflows. This paper
presents a BIM-enabled automation layer that extends a previously developed
surface-level heat-flux analysis framework by clarifying the IFC authoring
conditions and geometry-processing steps required to generate physically consistent, surface-resolved input features. Using IfcOpenShell, wall and window areas, outdoor-facing orientations, adjacency relationships, and zone floor areas are extracted directly from IFC geometry and assembled into machine-learning-ready inputs for an existing, validated XGBoost surrogate model. Through comparative case studies, the paper demonstrates that commonly used fa¸cade-level IFC authoring practices are structurally incompatible with surface-level analysis, as they prevent unambiguous association between envelope surfaces and thermal zones. In contrast, a zone-resolved IFC representation enables robust geometry-based linking and reliable extraction of surface-level features. Results show that, when these authoring conditions are satisfied, the extracted surface-level inputs preserve the dominant physical drivers of relative thermal performance. Surface-level rankings derived from the IFC-based workflow exhibit exact correspondence with a reference sensitivity-based analysis, even when geometric representations differ from earlier OpenStudio implementations. The proposed workflow reduces manual preprocessing effort and provides a reproducible pathway for integrating BIM data into data-driven retrofit analysis.