- D3.509 - Predicting Fungal and Bacterial Bioaerosol Concentrations Across Distinct Building Types: A Generalized Additive Model and Machine Learning Framework

Indoor bioaerosols are recognized environmental health hazards in public building environments. Under climate change, concerns regarding microbial exposure are increasing as indoor temperature and relative humidity (RH) conditions become more variable. However, microbial responses to indoor meteorological conditions differ markedly across facility types, limiting the effectiveness of uniform indoor air quality (IAQ) management strategies. We aimed to reclassify public multi-use facility types based on their sensitivity to indoor temperature and RH and to evaluate whether this facility-specific classification improves the prediction of airborne microbial concentrations.

Airborne bacteria and mold, along with indoor temperature and RH, were measured across 26 types of public multi-use facilities. Generalized Additive Mixed Models (GAMMs) were applied to identify statistically significant nonlinear associations between microbial concentrations and indoor meteorological conditions. Based on these associations, facilities were reclassified into four groups: sensitive to both temperature and RH, temperature only, RH only, or neither. Microbial concentrations were further categorized into safe, caution, and danger levels according to EU Good Manufacturing Practice (GMP) guidelines and Korean Ministry of Environment standards. Random Forest (RF) models were trained using the full dataset and reclassified subsets, and predictive performance was compared using F1 scores.

Microbial concentrations at caution or danger levels were most frequently observed during summer and fall. Reclassification of facilities based on meteorological sensitivity led to progressive improvements in predictive performance. The magnitude of improvement was greater for mold than for bacteria, reflecting the stronger environmental sensitivity of fungal aerosols to temperature and RH.

This integrated GAMM–RF framework provides a facility-specific, data-driven approach to improving IAQ assessment and microbial risk prediction in public multi-use buildings, supporting more targeted and effective indoor environmental health management strategies.