Machine learning algorithms for predicting air quality index: A case study in urban and industrial zones

In urban and industrial areas, the prediction of the air quality index (AQI) is important in order to control the air pollution and protect public health. The goal of this work is to enhance the AQI prediction by making use of the advanced machine learning (ML) and deep learning (DL) models capable of learning spatial and temporal dependencies. The main goal of this research is to examine the performance of the different ML and DL models such as Random Forest (RF), XGBoost, LSTM, Transformer and Temporal Graph Neural Networks (TGNN) for AQI prediction in urban and industrial zones. To capture the variability of data, a multi-source data collection approach is taken by using air quality data (PM2.5, PM10, SO2, NO2, CO, O3), weather data, satellite imagery, and IoT sensor data. The data were pre-processed and engineered in terms of temporal and spatial features and advanced models were used to predict AQI. And key metrics of RMSE, MAE and R2 were used to evaluate model performance. Results indicate that Transformer models achieved the best performance in urban areas, with an RMSE of 14.1 and R² of 0.89, because they can capture long-term temporal patterns. In industrial zones, TGNN models achieved an RMSE of 17.9 and an R² of 0.87 because they could capture spatial correlations and pollution dispersion. Both models exhibited high resilience to extreme pollution events and minimal performance degradation under missing data scenarios in robustness testing. We show that Transformer and TGNN models outperform traditional ML models by a large margin in AQI prediction, especially during high pollution episodes. The results are consistent with real-time air quality monitoring and dynamic policy making in urban and industrial environments. Future work should implement the models in other regions and improve data quality to increase applicability.