Evolution of PM 2.5 Forecasting: A Comprehensive Review of Statistical, Machine Learning, and Deep Learning Approaches

Air pollution, particularly fine particulate matter ($\mathrm{P M}_{2.5}$), poses a significant threat to human health and environmental sustainability. Accurate forecasting of$\mathbf{P M}_{2.5}$concentrations is crucial for early warning systems and effective air quality management. In recent years, machine learning (ML) and deep learning (DL) approaches have gained prominence due to their ability to capture complex nonlinear relationships in atmospheric data. This review provides a comprehensive analysis of state-of-the-art ML and DL models applied in$\mathbf{P M}_{2.5}$forecasting, including regression-based models, support vector machines, random forests, long short-term memory (LSTM) networks, convolutional neural networks (CNNs), and hybrid architectures. The paper discusses key input parameters such as meteorological factors, pollutant interactions, and spatiotemporal data sources used for model training. Furthermore, common evaluation metrics—such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of Determination ($\mathbf{R}^{2}$), and Index of Agreement (IA)—are analyzed to assess forecasting accuracy and model robustness. By synthesizing recent developments and highlighting comparative performance trends, this review aims to identify existing research gaps and provide insights into the selection of optimal models for$\mathbf{P M}_{2.5}$prediction.