Prediction of air pollutants in North-Western Nigeria: a comparison of LSTM and facebook prophet models

Abstract

Air pollution, driven by urbanization and globalization, is one of the major environmental problems facing society; it affects the physical environment and human health and, as a result, contributes significantly to increased morbidity and mortality. The dynamic, non-stationary nature of air-pollution time series makes air quality forecasting challenging. This has generated increasing interest in the application of advanced time-series and machine-learning models for air-quality forecasting. In this paper, two novel models are proposed: A Long Short-Term Memory (LSTM) and a Facebook Prophet Model (FPM), in which air pollutants were predicted. We used air pollution data from 10 features in the Northwestern states of Nigeria from 2018 through 2023. This research evaluated prediction results using RMSE, MAE, MAPE, and R². The study shows that while both models can effectively predict air pollution levels, the LSTM model had the lowest error rates for certain pollutants in designated states. At the same time, the Facebook Prophet Model displayed broader applicability across multiple pollutants and locations. The study also identified common regional factors contributing to air pollution, potentially tied to industrialization, population growth, and deforestation in northern Nigeria. The results suggest that, for comprehensive air quality monitoring, model selection should consider the specific pollutants and geographical location. Overall, although LSTM demonstrated superior performance for some pollutant-location combinations, the prophet model showed greater generalization across multiple pollutants and locations.