Air quality index AQI classification based on hybrid particle swarm and grey wolf optimization with ensemble machine learning model

Abstract

Accurate Air Quality Index (AQI) classification is essential for environmental surveillance and public health decision-making. Using a publicly available daily U.S. county-level dataset with six AQI categories (Good, Moderate, Unhealthy for Sensitive Groups, Unhealthy, Very Unhealthy, Hazardous), we conducted a comprehensive benchmarking study. Data preprocessing included missing-value imputation and class balancing via Synthetic Minority Over-sampling Technique (SMOTE). We trained and evaluated classical and deep models (Random Forest (RF), Extra Trees (ET), K-Nearest Neighbors (KNN), Naive Bayes (NB), Logistic Regression (LR), and a Multi-Layer Perceptron (MLP)) and assessed performance using cross-validation accuracy, test accuracy, macro-averaged recall, F1-score, and ROC-AUC. Ensemble methods (RF, ET) and the MLP consistently outperformed traditional baselines. RF achieved 99.3% test accuracy with perfect recall, F1-score, and ROC-AUC; MLP achieved 99.0% test accuracy. A stacking ensemble, optimized with a hybrid Particle Swarm–Grey Wolf Optimizer (PSO–GWO), delivered 99.99% test accuracy, 99.99% macro-averaged recall, and 1.0000 ROC-AUC. These findings demonstrate that combining ensemble learning with metaheuristic optimization can substantially enhance multi-class AQI classification performance and offer a practical path toward reliable, real-time air-quality assessment.