MACHINE LEARNING PREDICTION OF INDOOR PM₂.₅-BOUND DIBENZO[a,h]ANTHRACENE IN CHILDREN’S CHURCH FACILITIES IN SOUTHERN NIGERIA USING MICROCLIMATIC DATA

Abstract

Polycyclic aromatic hydrocarbons (PAHs) are hazardous semivolatile organic compounds frequently detected in indoor environments, several of which are designated by the United States Environmental Protection Agency (USEPA) as priority pollutants due to their carcinogenic potential. Dibenzo[a,h]anthracene (DahA), a high-molecular-weight PAH strongly associated with combustion-derived particulate matter, is classified by the International Agency for Research on Cancer (IARC) as probably carcinogenic to humans. This study applied supervised machine learning (ML) to predict indoor concentrations of PM₂.₅-bound DahA in children’s church facilities in Ugbowo, southern Nigeria, using microclimatic predictors. A total of 30 indoor air samples were collected from 5 locations across dry and wet seasons, and concentrations of the 16 USEPA priority PAHs were quantified by gas chromatography–mass spectrometry. Concurrent measurements of temperature, relative humidity, wind speed, atmospheric pressure, and dew point were obtained. Decision tree (DT), random forest (RF), linear regression (LR), and generalized linear model (GLM) algorithms were trained and evaluated using mean squared error (MSE), root mean squared error (RMSE), and the coefficient of determination (R²). Random Forest outperformed the other models (R² = 0.71), demonstrating superior capability in capturing nonlinear interactions between PM₂.₅ loading and microclimatic variables. While DahA prediction showed greater variability due to relatively low ambient concentrations, the Random Forest model effectively reproduced temporal and microenvironmental trends in DahA concentrations. The results demonstrate the utility of ensemble ML approaches for indoor PAH assessment and highlight the relevance of microclimatic controls for exposure mitigation in child-centric indoor environments.