Evaluation and Enhancement of an Indoor Air Quality Prediction Model for Infant Care Facilities Using Automated Relearning

Infants are particularly vulnerable to indoor air pollution due to their developing respiratory systems and prolonged time spent indoors. This study proposes a dynamic indoor air quality (IAQ) prediction model for daycare centers using automated machine learning (Auto ML) and monthly relearning. The model integrates real-time and historical data to address variability caused by occupant behavior, ventilation, and environmental conditions. A total of 446,611 observations were collected over 16 months from a two-story daycare center in South Korea, measuring CO₂, PM_2.5, PM₁₀, and TVOCs every 10 min. Among tested algorithms, ensemble learning methods (e.g., VotingEnsemble and XGBoost) showed superior performance. The model achieved predictive accuracies of 80%–89% for CO₂, 77%–98% for PM_2.5, 78%–97% for PM₁₀, and 70%–99% for TVOCs. Compared to prior studies focused on controlled environments or single-variable input, this model leverages diverse indoor–outdoor variables and continuous data accumulation, enabling real-time IAQ management. The approach is scalable to other sensitive facilities such as schools and healthcare centers. These findings demonstrate the potential of AI-based prediction frameworks for enhancing IAQ control strategies and protecting vulnerable populations.