An Explainable BIM-Machine Learning Framework for Predicting Energy Use, Cost, and Thermal Comfort in Residential Buildings in Bangladesh’s Hot-Humid Monsoon Climate

Early design decisions significantly influence the lifecycle energy consumption, cost, and occupant comfort of residential buildings. However, research integrating Building Information Modeling (BIM) with machine learning (ML) typically predicts these outcomes independently, employing opaque models that are primarily tested on institutional buildings in temperate or arid regions. Research on BIM in Bangladesh remains confined to single-outcome, non-ML energy simulations, leaving largely unexplored the potential for an interpretable, multi-outcome BIM-ML tool tailored for hot-humid monsoon residential design.

A parametric residential building model reflective of Bangladesh's hot-humid monsoon climate was constructed using Autodesk Revit, generating 100 design variants by adjusting the window-to-wall ratio, orientation, envelope thermal transmittance, and shading depth. For each variant, the annual energy consumption, estimated cost, and indoor thermal comfort are assessed using the ASHRAE 55 adaptive comfort index and evaluated through simulation with Autodesk Insight. The dataset was employed to train and compare Random Forest, XGBoost, and Artificial Neural Network models, with the highest-performing model subsequently interpreted through SHapley Additive exPlanations (SHAP) to quantify parameter influence and identify the optimal variant.

Random Forest emerged as the most effective model overall, attaining a test-set R² of 0.89 for energy consumption, 0.85 for cost, and 0.81 for thermal comfort. SHAP analysis identified the window-to-wall ratio and shading depth as the most significant predictors of energy consumption and comfort, while the envelope material exerted the greatest influence on cost. The top-performing variant, characterized by a 25% window-to-wall ratio, north-south orientation, and moderate shading, demonstrated an approximately 13% reduction in energy consumption and increased comfortable occupied hours by eight percentage points relative to the baseline, with an incremental cost of less than 4%.

The proposed framework offers a practical and interpretable BIM-ML decision support tool, designed to balance energy performance, cost, and occupant comfort during the initial phases of residential design. By establishing its feasibility within Bangladesh's hot-humid monsoon climate, a context often overlooked in existing BIM-ML research, the framework advances climate-responsive and cost-effective residential design practices across South Asia.