Interpretable predictive model-based architectural design strategies for mitigating reflected daylight glare in urban residential buildings

The spread of highly reflective facades aggravates the glare problem resulting from reflected daylight, which reduces urban residents’ visual comfort. Accordingly, this study proposes an explainable machine learning-based predictive model to quantitatively predict reflected daylight glare under various urban conditions and to interpret the primary factors influencing the glare. In the proposed model, various building and solar geometric conditions were generated via parametric simulation, and a dataset was built using feature engineering. The results of prediction using the XGBoost demonstrated that the model exhibited high performance (R² = 0.8970, RMSE = 0.0413, MAE = 0.0303) and faster evaluation (0.87 ms) compared to simulation. In addition, Shapley Additive exPlanations (SHAP) analysis was used to determine the causal relationships among key features. The analysis found that daylight entering angle (21.7%), glare façade solid angle (20.7%), solar altitude (13.4%), window height (8.8%), reflected daylight entering angle (8.5%), reflection angle (5.3%), and roughness (5.0%) were identified as key features, collectively accounting for 83.6% of the total reflected glare. Moreover, areas susceptible to reflected daylight glare were derived regarding key features, through which architectural design strategies were proposed for mitigating reflected daylight glare. The intensity of reflected daylight glare was significantly influenced by the concentration range of reflected light, depending on the position between the sun and the reflective surface, which can serve as quantitative evidence used for building design strategies. The findings of this study can also be used to control reflected daylight glare in urban façade design and to establish reflected light management policies.