Real-time Nowcasting of Urban Air Quality Index Based on LightGBM and Feature Engineering
TL;DR
The proposed LightGBM model can effectively achieve real-time AQI nowcasting for Hefei City and provides a scientific method for real-time air pollution monitoring and early warning.
Abstract
To improve the accuracy of real-time urban air quality index (AQI) nowcasting and address the insufficient rationality of feature engineering in machine learning models, this paper takes the daily air quality and meteorological monitoring data of Hefei City from January to November 2025 as the research object. A full-process framework of "core feature selection - multi-dimensional expansion - statistical optimization" is constructed for real-time AQI estimation. Combined with feature engineering and the LightGBM model, the realtime AQI nowcasting research is carried out. Firstly, PM10, PM2.5, and O3 are selected as core features through correlation analysis. Then 15 time-domain extended features are constructed. Feature selection and quantity sensitivity analysis are conducted using LightGBM feature importance. The results show that the core influencing factors of Hefei's AQI are PM10, PM2.5, and O3. The test set R 2 of the model with 3 core features reaches 0.872 and RMSE is 7.828. The proposed LightGBM model can effectively achieve real-time AQI nowcasting for Hefei City and provides a scientific method for real-time air pollution monitoring and early warning.