Machine Learning Based Air Quality Prediction for Smart Cities: A Comprehensive Review of Deep Learning Architectures, Hybrid Models, and Real-World Applications

Abstract

Pollution of air is one of the most critical environmental and public-health problems in urban regions and it is important to have predictive models to predict and give early warning. The literature survey of the paper covers all the machine learning and deep learning models used in air quality index (AQI) and concentration of individual pollutants forecasts in smart cities. In this work, we review systematically 20 recent papers that introduce various supervised learning algorithms such as Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU), Convolutional Neural Networks (CNN) and ensemble algorithms like XGBoost and Random Forests. The results show that the hybrid models (STGNN) are more accurate than the individual models in terms of predictive accuracy (R2>0.85, RMSE<15 μg/m3). The key results show that the models obtained by feature engineering of meteorological variables and advanced decomposition methods (EEMD-CEEMDAN) are robust. There remain challenges to be solved, including cross city generalization, efficiency of edge deployment and uncertainty quantification. The current methods are reviewed, the gaps in the current research are identified and future research directions including physics informed neural network learning and adaptive transfer learning for cross city generalization are proposed. The researchers and practitioners can use our research for developing smart environmental monitoring systems for sustainable cities.