Feature-Gated Spatio-Temporal Deep Learning With City Embedding And Attention-Based Bilstm For Air Quality Index Prediction
DOI:
https://doi.org/10.29284/rjgsay47Keywords:
Air Quality Index Prediction, Spatio-Temporal Deep Learning, Bidirectional Long Short-Term Memory (BiLSTM), Attention Mechanism, City Embedding, Feature Engineering, Multi-Task Learning, Urban Air Pollution Forecasting.Abstract
Air pollution is one of the biggest environmental and public health challenges faced by rapidly urbanizing areas, so accurate air quality prediction is crucial for environmental monitoring and policy making. This study employs a novel deep learning framework namely, Feature-Gated City-Embedded BiLSTM with Self-Attention (FGCE-BiLSTM) to predict Air Quality Index (AQI) robustly across multiple metropolitan cities. This model incorporates spatio-temporal feature learning and city-specific contextual representation to achieve improved regression-based AQI prediction and categorical air quality classification. The dataset used for this study is the 219,125 hourly observations collected from five great cities of India (Delhi, Mumbai, Chennai, Kolkata and Bangalore) featured on Kaggle platform. An exhaustive preprocessing pipeline is applied, which encompasses temporal feature extraction, missing value imputation, pollutant ratio construction, lagged-based generation of temporally-dependent features and rolling statistics for all pollutants. 67 engineered features are created in first place, and subsequently filtered against the correlation where a pair of variables would add mutual informative until an optimized feature set emerges from the cleaning process (hovering around 45 at best).
Sequential samples are generated from each city to capture temporal dependencies, and these are subsequently conditioned on city encodings in order to model spatial heterogeneity across urban environments. The feature-gating mechanism dynamically adjusts the importance of the features before temporal modeling based on stacked Bidirectional Long Short-Term Memory (BiLSTM) layers. Then a self-attention mechanism can be utilized to emphasize the most impactful temporal patterns. Based on two-output strategy, the framework performs AQI regression and air quality category classification simultaneously. Because of large class imbalance the Good and Satisfactory classes were ignored, and four representative classes (Moderate, Poor, Severe and Very Poor) were used to train the model. The predictive ability was shown experimentally both in regression (R², MAE, RMSE) and classification tasks (Accuracy, Precision, Recall and F1-score), as well as confusion matrix and ROC curves. Overall, the findings demonstrate that with spatio-temporal modeling capability, the proposed FGCE-BiLSTM model can learn complex pollution patterns and greatly enhances city-wide prediction reliability. This framework presents a scalable and interpretable method for intelligent air quality monitoring, facilitating data-driven environmental decision-making in urban ecosystems.
References
[1] G. Ravindiran, G. Hayder, K. Kanagarathinam, A. Alagumalai, and C. Sonne, “Air quality prediction by machine learning models: A predictive study on the Indian coastal city of Visakhapatnam,” Chemosphere, vol. 338, Art. no. 139518, 2023. https://doi.org/10.1016/j.chemosphere.2023.139518
[2] Y. C. Liang, Y. Maimury, A.H.L. Chen, and J.R.C. Juarez, “Machine learning-based prediction of air quality,” Applied Sciences, vol. 10, no. 24, Art. no. 9151, 2020. https://doi.org/10.3390/app10249151
[3] J. Zhang, and S. Li, “Air quality index forecast in Beijing based on CNN–LSTM multi-model,” Chemosphere, vol. 308, Art. no. 136180, 2022. https://doi.org/10.1016/j.chemosphere.2022.136180
[4] J. Duan, Y. Gong, J. Luo, and Z. Zhao, “Air-quality prediction based on the ARIMA-CNN-LSTM combination model optimized by dung beetle optimizer,” Scientific Reports, vol. 13, Art. no. 12127, 2023. https://doi.org/10.1038/s41598-023-36620-4
[5] T. Xu, Y. Tian, X. Cai, C. Wu, and Z. Mian, “Air quality forecasting and rating based on machine learning algorithm and cumulative logit model: an empirical study for Lanzhou city of China,” Environment, Development and Sustainability, 2025. https://doi.org/10.1007/s10668-024-05792-y
[6] H. Alrashidi, F.N. Sibai, A. Abonamah, M. Alrashidi, and A. Alsaber, “PM2.5: Air Quality Index Prediction Using Machine Learning: Evidence from Kuwait’s Air Quality Monitoring Stations,” Sustainability, vol. 17, no. 20, Art. no. 9136, 2025. https://doi.org/10.3390/su17209136
[7] T. M. T. Lei, J. Cai, W.H. Cheng, T.A. Kurniawan, A.H. Molla, M.S.M. Nadzir, S.S.K. Kong, and L.W.A. Chen, “Application of Deep Learning Techniques for Air Quality Prediction: A Case Study in Macau,” Processes, vol. 13, no. 5, Art. no. 1507, 2025. https://doi.org/10.3390/pr13051507
[8] G. Ma, K. Xu, Y, Zhang, L. Zhang, and Z. Chen, “A novel deep learning model for air quality index prediction integrating time series decomposition and intelligent optimization,” Results in Engineering, vol. 27, Art. no. 106078, 2025. https://doi.org/10.1016/j.rineng.2025.106078
[9] A. Samad, S. Garuda, U. Vogt, and B. Yang, “Air pollution prediction using machine learning techniques – An approach to replace existing monitoring stations with virtual monitoring stations,” Atmospheric Environment, vol. 310, Art. no. 119987, Oct. 2023.
https://doi.org/10.1016/j.atmosenv.2023.119987
[10] J. Wang, X. Li, L. Jin, J, Li, Q. Sun, and H. Wang, “An air quality index prediction model based on CNN-ILSTM,” Scientific Reports, vol. 12, Art. no. 8373, 2022. https://doi.org/10.1038/s41598-022-12355-6
[11] J. Chang, J. Hou, H. Gong, and Y. Sun, “Optimization of CNN-LSTM Air Quality Prediction Based on the POA Algorithm,” Sustainability, vol. 17(12), Art. no. 5347, 2025. https://doi.org/10.3390/su17125347
[12] Y. Hu, Y. Ding, and W. Jiang, “Geographically Aware Air Quality Prediction Through CNN-LSTM-KAN Hybrid Modeling with Climatic and Topographic Differentiation,” Atmosphere, vol. 16(5), Art. no. 513, 2025. https://doi.org/10.3390/atmos16050513
[13] E. Mohapatra, R. Das, and S. Rath, “Deep learning-based AQI forecasting: a CNN LSTM model with visual insights from SHAP LIME and PDP,” Discover Applied Sciences, vol. 7, Art. no. 1326, 2025. https://doi.org/10.1007/s42452-025-07845-x
[14] R. Janarthanan, P. Partheeban, K. Somasundaram, and P.N. Elamparithi, “A deep learning approach for prediction of air quality index in a metropolitan city,” Sustainable Cities and Society, vol. 67, Art. no. 102720, 2021. https://doi.org/10.1016/j.scs.2021.102720
[15] X. Wu, X. Gu, and K. W. See, “ADNNet: Attention-based deep neural network for air quality index prediction,” Expert Systems with Applications, vol. 250, Art. no. 125128, 2024.
https://doi.org/10.1016/j.eswa.2024.125128
[16] J. Wang, J. Liu, X. Wang, T. Wang, and Q. Sun, “An air quality prediction model based on CNN-BiNLSTM-Attention,” Environment, Development and Sustainability, vol. 27, pp. 24705 – 24720, 2025. https://doi.org/10.1007/s10668-021-02102-8
[17] Z. Zhang, S. Zhang, X. Zhao, L. Chen, and J. Yao, “Temporal difference-based graph Transformer networks for PM2.5 prediction: A Case Study in China,” Frontiers in Environmental Science, vol. 10, 2022. https://doi.org/10.3389/fenvs.2022.924986
[18] X. Liu, K. Su, S. Wang, and K.H. Ghazali, “Intelligent prediction of air quality index based on the Transformer-BiLSTM model,” Scientific Reports, vol. 15, Art. no. 41838, 2025. https://doi.org/10.1038/s41598-025-25865-w
[19] Y. Ye, H. Zhang, J. Li, and X. Chen, “A graph neural network and Transformer-based model for PM2.5 concentration prediction,” Environmental Modelling & Software, vol. 181, Art. no. 106501, 2025. https://doi.org/10.1016/j.envsoft.2025.106501
[20] R. Pan, T. Liu, and L. Ma, “A Graph Attention Recurrent Neural Network Model for PM2.5 Prediction: A Case Study in China from 2015 to 2022,” Atmosphere, vol. 15(7), Art. no. 799, 2024. https://doi.org/10.3390/atmos15070799
[21] Y. Li, R. Yu, C. Shahabi, and Y. Liu, “Diffusion convolutional recurrent neural network: Data-driven traffic forecasting,” International Conference on Learning Representations (ICLR), 2020. https://arxiv.org/abs/1707.01926
[22] Y. Huang, F. Han, and Q. Feng, “A novel model for predicting PM2.5 concentrations utilizing graph convolutional networks and Transformer,” IEEE Access, vol. 13, pp. 30613–30622, 2025. https://doi.org/10.1109/ACCESS.2025.354177
[23] J. Cai, X. Fu, B. Su, and H. Fang, “Air quality prediction based on spatio-temporal feature fusion over graphs,” Processes, vol. 13(11), Art. no.3586, 2025. https://doi.org/10.3390/pr13113586
[24] Dataset source: https://www.kaggle.com/datasets/sathvikisikella/cleaned-air-quality-india/data
Downloads
Published
Issue
Section
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
This work is licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
