Semiparametric hybrid air quality forecasting using prophet ensemble and feature selection

Area of Science:

• Environmental Science
• Data Science
• Public Health

Background:

• Accurate prediction of fine particulate matter (PM2.5) concentrations is essential for effective air quality management and public health planning.
• High-dimensional environmental datasets, characterized by numerous lagged variables and complex pollutant interactions, pose significant challenges to traditional forecasting models.
• Existing methods may struggle to maintain performance and interpretability when faced with the complexity of real-world air quality data.

Purpose of the Study:

• To develop and evaluate a novel hybrid forecasting framework for PM2.5 concentrations.
• To address the challenges posed by high-dimensional environmental data in air pollution forecasting.
• To enhance the accuracy and interpretability of air quality predictions.

Main Methods:

• A hybrid forecasting framework was developed, integrating the Prophet time series model with ensemble-based machine learning techniques for residual modeling.
• Ensemble methods including bagging, boosting, stacking, and voting regressors were employed to model Prophet's residuals.
• Feature selection techniques such as Lasso, Shap, and Recursive Feature Elimination were applied to manage high-dimensionality and identify influential features.

Main Results:

• The hybrid Prophet model combined with boosting, Lasso, and Shap demonstrated the lowest prediction errors (RMSE) compared to other ensemble residual models and standalone machine learning models (XGBoost, SVR, LSTM).
• The proposed framework effectively handled high-dimensional simulated and real-world air quality data from Seoul, South Korea.
• The selected feature selection methods enhanced the interpretability of the influential factors driving PM2.5 concentrations.

Conclusions:

• The hybrid Prophet with boosting ensemble model, enhanced by Lasso and Shap feature selection, offers a statistically sound, interpretable, and scalable solution for forecasting air pollution in high-dimensional settings.
• This approach significantly improves upon existing methods for PM2.5 prediction, providing a valuable tool for environmental and public health management.
• The study highlights the efficacy of combining time series decomposition with advanced machine learning for complex environmental forecasting tasks.