Enhanced Air Quality Prediction Using AI: A Comparative Study of GRU, CNN, and XGBoost Models

Weather monitoring is vital due to environmental changes and rising air pollution, which affects health and lifestyles. Accurate air quality prediction models are essential yet challenging due to complex weather-pollution interactions. This study employs explainable deep learning and machine learning techniques—GRU, CNN, and XGBoost—on a custom dataset of 100,000 samples with 15 features, including PM2.5, PM10, humidity, and temperature. Using SHAP for interpretability, the GRU model outperforms others with 98.56% accuracy, 98.43% Recall, and 98.52% True Positive Rate. Temperature, humidity, gases, and pressure are key variables influencing predictions. The proposed model achieves high mAP and precision, surpassing existing methods and demonstrating effective real-time forecasting under diverse weather conditions.