<![CDATA[Accurate prediction of Air Quality Index (AQI) is critical for mitigating public health risks associated with urban air pollution. This study presents an empirical analysis of PM2.5 concentration forecasting in Beijing using advanced machine learning algorithms, integrating high-resolution atmospheric data and meteorological variables. A multi-stage pipeline was implemented, including data preprocessing, feature selection, and model training with Random Forest, Gradient Boosting, Support Vector Regression, and Long Short-Term Memory (LSTM) networks. Predictive performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and R², while spatial and temporal fidelity was assessed across districts, diurnal cycles, and seasonal periods. LSTM models consistently achieved superior accuracy, capturing short-term pollution spikes, seasonal variability, and spatial heterogeneity, whereas ensemble methods provided stable baseline predictions with moderate sensitivity to extreme events. Sensitivity analysis identified wind speed, humidity, and neighboring PM2.5 measurements as key predictors. The results demonstrate that integrating recurrent neural networks with ensemble approaches enables reliable, operationally relevant AQI forecasts, offering both theoretical validation of sequential modeling for urban air quality and practical guidance for environmental monitoring, public health interventions, and city-level policy implementation. ]]>
Copyrights © 2026
