Reduced Bayesian Optimized Stacked Regressor (RBOSR): A highly efficient stacked approach for improved air pollution prediction

Over the past few years, stacked generalization has emerged as a promising approach for predicting the concentration of PM2.5, but its use is often associated with reduced efficiency due to computational demands. In this study, we propose a new approach called Reduced Bayesian Optimized Stacked Regressor (RBOSR) that optimizes both the performance and efficiency of the PM2.5 stacked model. RBOSR incorporates Bayesian optimization for hyperparameter tuning, ensemble-based feature selection, dimensionality reduction via single-link hierarchical clustering, and recursive base estimator eliminations. Results: The RBOSR model is significantly more efficient when compared to the original stacked model, with 5.7 times shorter training time. Additionally, the RBOSR model outperformed the unreduced stacked model with an R2 value of 0.91 and RMSE of 26.46, resulting in a 3.5% improvement. Compared to other PM2.5 stacked models proposed in recent studies, the RBOSR model demonstrates superior efficiency, with up to 47 times shorter training time. While the RBOSR approach has been developed specifically for PM2.5 prediction, it has the potential for broader applications in other regression problems. Although the approach has not yet been applied to other datasets, future work could explore its applicability to a broader range of datasets and the development of more efficient strategies for optimizing base estimator selection.

Automated summary

RBOSR is a new approach that improves the performance and efficiency of the PM2.5 stacked model. It significantly reduces training time and outperforms the original model.