Feature Selection in Ozone Feature Space Impacts Performance in Gradient Boosting, Random Forest, Xgboost and Adaptive Boosting Regressors

This study investigated the efficacy of feature selection methods in enhancing ozone prediction through regression models, utilizing the “AQBench” dataset, abundant with air quality indicators and environmental variables. We focused on Adaptive Boosting, Gradient Boost Regressor, Random Forest Regressor, and XGBoost Regressor, assessing the impact of various feature selection strategies—such as Feature Shuffling, Random Forest Importance, and Step-Forward Feature Selection—on model performance. Our findings highlighted that the XGBoost Regressor stood out for its accuracy and generalizability across different feature selection methods, emphasizing the critical role of targeted feature selection in environmental predictive modeling. This research contributes to integrating machine learning techniques in environmental science, providing insights that could influence public health and policy decisions. It underscores the potential of computational approaches to enhance air quality forecasts, paving the way for future investigations into additional models and datasets to refine environmental predictive analytics.