Comparative Analysis of Tree-Based Ensemble Methods for Air Quality Predictions in Singapore Using Socioeconomic Attributes

Air quality is a significant public health issue, as poor air quality presents considerable health risks and environmental challenges. With all previous research in air quality prediction, the integration of socioeconomic factors is often neglected. This study addresses this gap by evaluating tree-based ensemble learning methods for predicting air quality with socioeconomic data. The main objective is to compare the predictive power, stability, and computational efficiency of bagging, boosting, stacking, and multi-layer stacking models in forecasting Singapore's Pollutant Standards Index (PSI). Utilizing a dataset of 34 socioeconomic attributes, we employed 10-fold cross-validation and conducted thorough hyperparameter tuning and statistical tests to validate the performance differences among the models. Our findings indicate that tree-based ensemble models are effective in predicting PSI, with Random Forest models showcasing superior accuracy and consistency. Boosting models like LightGBM exhibited better computational efficiency and stability, whereas stacking models underperformed, likely due to challenges with the small, high-dimensional dataset that led to overfitting. Statistical tests confirmed significant performance differences among the models. These results underscore the potential benefits of incorporating socioeconomic factors into air quality modeling and highlight the importance of selecting appropriate models based on dataset characteristics to improve forecasting accuracy and public health outcomes.