Determining Minimum Monitoring Period for Predicting Occupant Behaviors Related to Manual Control of Windows in Residential Buildings

Occupant behavior related to the manual control of indoor environmental systems, particularly window operation, plays a significant role in energy use and indoor environmental quality. Manual window opening for natural ventilation strongly affects both indoor thermal conditions and air quality. Previous studies have shown that machine learning models can predict occupant behavior more accurately than logistic regression models, with key influencing factors including indoor and outdoor environmental parameters. This study analyzed the effects of monitoring period and temporal resolution on the accuracy of predicting occupants' window-opening behavior to support occupant-centric control of residential ventilation and HVAC systems. Field-monitored data collected from apartment samples over more than six months were analyzed, and Random Forest and k-Nearest Neighbors (KNN) models were used to predict manual window control using environmental factors (e.g., temperature, CO2, PM). The results showed that a 1-week period and a one-hour resolution were required for accurate prediction. These findings suggest that short-term monitoring is sufficient for implementing occupant-centric predictive control of residential ventilation and HVAC systems.