- Research Article
- 10.1016/j.jobe.2025.115046
Data imputation methods for missing U-values of building envelopes in building performance database
- Jan 01, 2026
- Journal of Building Engineering
- Wenzhuo Li + 4 more +4
Accurate thermal transmittance (U-value) data for building envelopes is crucial for planning effective retrofits and modeling building energy performance. However, missing U-values in existing building stock datasets present a significant barrier to large-scale decarbonization efforts. This study evaluates six data imputation methods for addressing missing U-values, comparing traditional approaches (mean/median imputation) with advanced techniques including multiple imputation via chained equations (MICE) with Bayesian ridge regression (BRR), k-nearest neighbours (KNN), random forest (RF), and LightGBM. The Energy Performance Certificates (EPCs) data from 112,125 domestic properties in London’s Barnet borough is used as a case study. The evaluation indicators are imputation errors and execution time. Different methods are further evaluated through sensitivity analysis in EnergyPlus. U-values are varied by each imputer’s error magnitude, so as to assess the impacts of uncertainty in imputed values on energy use and overheating predictions. Results demonstrate that MICE with LightGBM minimises uncertainty in building performance predictions while maintaining reasonable execution time, offering a robust solution for improving building stock data usability. • Investigated six imputation methods on 112K+ EPC records from Barnet, London. • MICE withLightGBM yields lowest imputation errors for missing U-values. • LightGBM imputation reduces uncertainty in building energy performance. • Sensitivity analysis in EnergyPlus links imputation choice to energy metrics. • Robust, efficient imputation aids accurate retrofit planning and modeling.
Read more