The study of indoor particulate matter in office buildings based on artificial intelligence

Abstract

The necessity of supplying proper indoor air quality in workplaces to provide the principles of a healthy and productive labor force and avoid negative outcomes is a known fact. This study assessed particulate matter (PM) concentrations in office buildings of governmental organizations across five regions in Tehran over four seasons (2018–2019) to model annual indoor PM patterns using machine learning. PM concentrations, including PM₁, PM_2.5, PM₁₀, and Total Particulate Matter (TPM), were categorized using ensemble modeling techniques such as Linear Regression, Random Forest, Gradient Boosting, XGBoost, CatBoost, Support Vector Regression, and K-nearest neighbors. Key air quality parameters measured were CO2 (784 ppm), SO2 (0.114 μg/m³), PM_2.5 (4.604 μg/m³), temperature (24.8 °C), and relative humidity (21.16%). While most parameters met guidelines, PM₁₀ levels (97.5 μg/m³) exceeded WHO standards and relative humidity was below recommended levels, highlighting areas for improvement. PM_2.5 and PM₁₀ showed the strongest positive correlation (p value = 0.0001) and similar seasonal trends, with higher concentrations in autumn and summer and lower levels in spring and winter. The southern region exhibited consistently higher PM concentrations, while no significant changes were noted in the East or West. Among the models, CatBoost performed best in predicting air quality. The study suggests that indoor PM levels are influenced by psychrometric conditions and building location, providing valuable insights for improving air quality and occupant health.