A study of subjective evaluation factors regarding visual effects of daylight in offices using machine learning

Daylight has a positive impact on the health, pleasure, and productivity of office workers. Analyzing how environmental parameters affect daylight's visual effects can help effectively evaluate daylight sensation and assist architects in creating more appropriate daylight environments. The study explores the impact of different daylight environments on occupants' perception via extensive controlled daylight experiments in Harbin, China. Based on the experimental results, we find two common factors in the visual effects of daylight through exploratory factor analysis and confirmatory factor analysis. Take the evaluation factors (daylight satisfaction and daylight complexity) with the highest factor loading coefficient in each common factor as an example, construct a predictive model based on the XGBoost algorithm and validate its performance. Then, the SHAP (SHapley Additive exPlanations) method is used to analyze the influence mechanism of feature parameters and their interaction combinations on the visual effects of daylight in office spaces. The results indicate that the feature parameter L_task/L_m significantly impacts daylight satisfaction and complexity. The interaction combination between feature parameters L_task/L_m and L_task significantly influences daylight satisfaction evaluation. In contrast, the combination of L_med and Daylight Glare Probability (DGP) contributes the most to daylight complexity. The study addresses theoretical and technical gaps in the visual effects study of daylight through questionnaire assessment and explanatory analysis, enhancing the precision and scientific rigor of daylighting design.