Machine learning and deep learning approaches for PM2.5 prediction: a study on urban air quality in Jaipur, India

Abstract

Air pollution poses a significant threat to public health and the environment. Understanding and predicting PM2.5 concentrations are crucial for implementing effective mitigation strategies. The present study aimed to predict PM2.5 in Jaipur city by utilizing air pollutants and meteorological parameters from 2019–2023 through five machine learning approaches: multiple linear regression (MLR), support vector regression (SVR), artificial neural network (ANN), random forest (RF), and k-nearest neighbors (KNN), Gated Recurrent Units (GRU) and Convolutional Neural Network (CNN). A total of 39,645 data points were pre-processed and analyzed, including multicollinearity tests, before model training. The result showed that all air pollutants exceeded World Health Organization (WHO) (2021) guidelines. An increasing trend of PM2.5 was noted with NO2, SO2, and NH3. The sensitivity analysis revealed that SO2 and O3 had a greater sensitivity in changes in PM2.5 concentrations. The maximum correlation of PM2.5 was observed with NO2. The maximum combined standardized effect of multiple parameters was noted for relative humidity and temperature on PM2.5 concentration. Among the models, CNN demonstrated superior performance with high prediction accuracy (R2 = 0.98) and low error, outperforming other models. The model ranking was CNN > ANN > KNN > RF > GRU > MLR. By utilizing machine learning approaches this study highlights the predictive capabilities of these models and provides actionable insights for policymakers and stakeholders in mitigating the harmful impacts of air pollution on public health and environmental sustainability. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024.