Prediction of BTEX concentrations in the air of Southern East Azerbaijan province, Iran using ensemble machine learning and feature analysis

• Novel ML integrates PM and environmental data for precise BTEX pollution prediction. • Dual-layer analysis reveals feature importance in base models and meta-learner. • Advanced ML boosts air BTEX prediction accuracy using multi-source data. BTEX (Benzene, Toluene, Ethylbenzene, and Xylene) compounds are prominent air pollutants with severe implications for human health. Prolonged exposure to these volatile organic compounds (VOCs) has been associated with respiratory problems, cancer, and neurological disorders. Consequently, accurate prediction of their concentrations is vital for safeguarding public health and ensuring environmental safety. In this study, we introduce the MLs (XGBRegressor, AdaBoostRegressor, ExtraTreesRegressor, and CatBoost)-Stacked-Extra Trees ensemble, an innovative machine learning approach to predict BTEX concentrations. The initial model selection process employed the LazyRegressor library, which efficiently evaluates a wide array of regression models and provides essential performance metrics. Based on R-squared values, the top-performing models identified were XGBRegressor, AdaBoostRegressor, and ExtraTreesRegressor. To further optimize the stacking ensemble, CatBoost, a high-performing model not included in LazyRegressor, was incorporated. A thorough feature analysis identified key predictors influencing BTEX concentrations, including PM 10 , PM 2.5 , humidity, temperature, wind speed, and UV index. Additionally, the contributions of each model within the ensemble were assessed, highlighting the advantages of integrating predictions from multiple models to enhance accuracy. Our findings indicate that the MLs-Stacked-Extra Trees ensemble significantly outperforms individual models, achieving R² values of 1.0 and 0.998 for training and testing datasets, respectively. This research underscores the potential of advanced machine learning techniques to monitor air quality and guide policy decisions aimed at mitigating health risks associated with VOCs exposure.