MACHINE LEARNING PREDICTION OF PM2.5 IN LAGOS USING EMBEDDED REAL-TIME ENVIRONMENTAL MONITORING

Abstract

This study aims to develop and evaluate an embedded system for real-time monitoring of PM2.5 and meteorological variables, with the goal of improving machine learning predictions of particulate matter concentrations in Lagos, Nigeria. Given the detrimental health effects of PM2.5, understanding its interaction with environmental factors is crucial for effective air quality control.  Over two years (2021–2023), our innovative, custom-developed sensor system was deployed in Akoka, Lagos, to continuously and autonomously collected temperature, humidity, wind speed, atmospheric pressure, and PM2.5 data at two-minute intervals. Leveraging this robust data set, three machine learning algorithms: Random Forest (RF), Gradient Boosting (GB), and Support Vector Machine (SVM), were systematically evaluated for PM2.5 forecasting using R² and RMSE as performance metrics. The Random Forest model demonstrated the best performance (R² = 0.77; RMSE = 10.84 µg/m³), indicating high predictive capacity. Feature importance analysis revealed a limited impact of meteorological variables compared to unmeasured emission sources. This work demonstrates the feasibility of embedded real-time monitoring integrated with ML for urban air quality forecasting, supporting improved policy and public health strategies in rapidly urbanizing regions.