Desenvolvimento de modelos globais para predição de insatisfação com a qualidade ambiental interior em edificações: uma análise em salas de aula universitárias

Abstract

Indoor Environmental Quality (IEQ) is essential to ensure healthy, comfortable, and productive environments, especially in places such as classrooms, where occupancy rates are high and people remain for long periods. Traditionally, its assessment focuses on indoor air quality and on thermal, visual, and acoustic comfort. However, users' perception involves a broader set of factors, including biophilia and outdoor views, cleanliness, maintenance, furniture, layout, and personal control. In this context, this research aimed to develop global models to estimate the predicted percentage of dissatisfaction with Indoor Environmental Quality in university classrooms, using five predictive approaches: Multiple Linear Regression (MLR), Artificial Neural Networks (ANN), Random Forest (RF), Gradient Boosting (GB), and Partial Least Squares (PLS). The motivation stems from the lack of models that integrate tangible (objective) and intangible (subjective) variables, both of which are relevant to occupants' experience. The results revealed moderate statistical performance, with coefficients of determination (R²) ranging from 0.534 to 0.582, a level consistent with similar studies in the field. The Random Forest (RF) model achieved the best overall performance (R² = 0.582), followed by Multiple Linear Regression (R² = 0.552), Gradient Boosting (R² = 0.549), Artificial Neural Networks (R² = 0.541), and Partial Least Squares (R² = 0.534). It was observed that frequently overlooked factors, such as layout, personal control, furniture, and maintenance, showed significant influence on dissatisfaction, with convergence across models regarding the relevance of these predictors. Furthermore, non-linear approaches demonstrated a greater capacity to capture complex relationships among variables, although linear regression performed competitively, indicating that linear relationships can explain a substantial portion of IEQ variability. The joint integration of the physical and perceptual domains is a promising strategy for guiding the planning, operation, and retrofit of educational buildings, with emphasis on interventions targeting layout, furniture, and personal control, factors that emerged as the most critical for reducing student dissatisfaction.