AI-Driven Modeling of Window Opening Behavior in Kindergarten Classrooms: A Case Study during the Transitional Season in the Cold Region of China

With increasing awareness of energy conservation and environmental protection, optimizing indoor air quality and energy consumption through rational control of window opening behavior (WOB) has become a crucial issue in building design and environmental management. However, existing research primarily focuses on buildings for adults, with relatively few studies on buildings for children, particularly those used by children aged 3–6. Moreover, previous studies often overlook the impact of functional differences between buildings on occupant behavior patterns. This study focuses on a kindergarten and proposes an event-based method for analyzing and modeling WOB. The results show that events such as arrival, class, and departure are associated with higher frequencies of window opening (exceeding 50%), whereas events such as dietary activity, indoor/outdoor activity, sleep, and tidying exhibit lower probabilities. WOB is more sensitive to indoor air quality during events with higher student activity (e.g., class, dietary activity, and indoor activity), resulting in more frequent ventilation. In terms of modeling, the Random Forest (RF) algorithm achieved higher prediction accuracy than Logistic Regression (LR) and Support Vector Machine (SVM). To reduce the complexity associated with multi-model integration, a stacking model was introduced, further enhancing predictive performance. Finally, the generalizability of the proposed method was validated using office building data from the ASHRAE occupant behavior database, achieving a maximum accuracy improvement of 3.87%. This study presents a novel approach for modeling WOB in functional buildings such as kindergartens and provides theoretical support for energy efficiency optimization and indoor air quality management.