Machine Learning-Based Building Energy Performance Prediction and Sustainable Architectural Design Optimization Using Building Information Modeling (BIM)

Abstract

Building energy performance prediction is essential for sustainable architectural design and the reduction of operational energy consumption. This study proposes a Building Information Modeling (BIM)-integrated machine learning framework for predicting annual building energy use intensity (EUI) and optimizing architectural design parameters. A parametric building model is linked to EnergyPlus simulation, while Random Forest Regression, Gradient Boosting Regression, and Artificial Neural Network (ANN) Regression are employed to estimate annual EUI. A multi-objective optimization procedure is subsequently applied to identify energy-efficient architectural alternatives. For demonstration, a simulated dataset comprising 1,000 building design alternatives was generated. The illustrative evaluation results show that the Gradient Boosting model achieved the highest coefficient of determination (R²=0.98), with an RMSE of 2.9 kWh/m²/year and an MAE of 2.1 kWh/m²/year. The hypothetical optimization reduced annual EUI from 142.6 kWh/m²/year for the baseline design to 105.8 kWh/m²/year for the optimized alternative, representing a reduction of 25.8%. The optimized configuration was assumed to incorporate improved glazing properties, suitable external shading, a reduced window-to-wall ratio, and enhanced envelope thermal performance. These quantitative results are illustrative and are not based on verified experimental measurements or actual simulation outputs. Nevertheless, the proposed framework demonstrates the potential of integrating BIM, building-energy simulation, machine learning, explainable artificial intelligence, and multi-objective optimization to support evidence-based, energy-efficient, and sustainable architectural design decisions.

.