DUAL-STREAM MODEL OF LONG SHORT-TERM MEMORY FOR AIR QUALITY PREDICTION USING STATION AND WEATHER DATA

Ivan Rudavskyi, Halyna Klym

Abstract


Background. Air pollution remains a critical environmental and public health issue, requiring accurate and reliable forecasting methods to support decision-making and mitigation strategies. With the increasing availability of environmental data from monitoring stations and meteorological sources, advanced data-driven approaches have become essential for modeling complex air quality dynamics. The development of robust deep learning architectures capable of capturing both temporal dependencies and external environmental influences is of significant importance.

Methods. This study proposes a Dual-Stream Long Short-Term Memory (LSTM) model for air quality forecasting, which processes station-based and meteorological data through two parallel streams. A preprocessing pipeline is applied, including K-nearest neighbors (KNN) imputation for handling missing values and Z-score normalization to standardize input features. The outputs of the LSTM streams are integrated using an attention-based fusion mechanism, which adaptively assigns importance to each data source. The model performance is compared with several baseline approaches.

Results and Discussion. Experimental results demonstrate that the proposed model achieves improved prediction accuracy and stability compared to traditional machine learning methods and single-factor models. The attention-based fusion mechanism effectively captures the dynamic relationships between station data and meteorological conditions, allowing the model to adapt to varying environmental influences. Evaluation using the Index of Agreement, coefficient of determination, and Root Mean Square Error confirms the superiority of the proposed approach in modeling complex air quality patterns.

Conclusion. The proposed Dual-Stream LSTM framework provides an effective solution for integrating heterogeneous environmental data in air quality forecasting tasks. By combining structured preprocessing techniques with an adaptive fusion strategy, the model enhances prediction performance and robustness. The results highlight the potential of multi-source deep learning approaches for environmental monitoring and can be extended to other time-series forecasting applications.

Keywords: air quality prediction; LSTM; dual-stream model; KNN imputation; Z-score; sensor.


Full Text:

PDF

References


[1] Kotlia, P., Pant, J., & Lohani, M. C. (2025, September). Design and Implementation of Statistical Time Series Forecasting Model for Air Quality Assessment. In 2025 6th International Conference on Electronics and Sustainable Communication Systems (ICESC) (pp. 913-918). IEEE. https://doi.org/10.1109/icesc65114.2025.11212556

[2] Pandian, E., et al. (2025). Air quality forecasting for predictive analysis using machine learning. In Proceedings of the 9th International Conference on Inventive Systems and Control (ICISC) (pp. 187–192). IEEE. https://doi.org/10.1109/icisc65841.2025.11188992

[3] Potey, S., et al. (2025). Air quality prediction using machine learning: A comprehensive review. In Proceedings of the International Conference on Cognitive Computing in Engineering, Communications, Sciences and Biomedical Health Informatics (IC3ECSBHI) (pp. 1089–1093). IEEE. https://doi.org/10.1109/ic3ecsbhi63591.2025.10991003

[4] Borah, J., et al. (2024). Timezone-aware auto-regressive long short-term memory model for multipollutant prediction. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 1–9. https://doi.org/10.1109/tsmc.2024.3463960

[5] Alani, N. H. S., Chand, P., & Al-Rawi, M. (2025). A Two-Stage Machine Learning Framework for Air Quality Prediction in Hamilton, New Zealand. Environments, 12(9), 336. https://doi.org/10.3390/environments12090336

[6] Çelikten, H. (2026). A Multi-Output Deep Learning Framework for Simultaneous Forecasting of PM10 and Air Quality Index in High-Altitude Basins: A Case Study of Igdir, Türkiye. Sustainability, 18(8), 3883. https://doi.org/10.3390/su18083883

[7] Rudavskyi, I., & Klym, H. (2026). Air quality prediction based on long short-term memory prediction model. Measuring Equipment and Metrology, 87(1). https://doi.org/10.23939/istcmtm2026.01.026

[8] Zhang, X., Sun, Z., Zhou, Z., Jamali, S., & Liu, Y. (2022). Analysis and Dynamic Monitoring of Indoor Air Quality Based on Laser-Induced Breakdown Spectroscopy and Machine Learning. Chemosensors, 10(7), 259. https://doi.org/10.3390/chemosensors10070259

[9] Taştan, M. (2025). Machine Learning–Based Calibration and Performance Evaluation of Low-Cost Internet of Things Air Quality Sensors. Sensors, 25(10), 3183. https://doi.org/10.3390/s25103183

[10] Okorn, K., & Hannigan, M. (2021). Improving Air Pollutant Metal Oxide Sensor Quantification Practices through: An Exploration of Sensor Signal Normalization, Multi-Sensor and Universal Calibration Model Generation, and Physical Factors Such as Co-Location Duration and Sensor Age. Atmosphere, 12(5), 645. https://doi.org/10.3390/atmos12050645

[11] Chen, H., Guan, M., & Li, H. (2021). Air quality prediction based on integrated dual LSTM model. IEEE Access, 9, 93285–93297. https://doi.org/10.23939/acps2025.01.001

[12] Ma, X., Chen, T., Ge, R., Xv, F., Cui, C., & Li, J. (2023). Prediction of PM2.5 Concentration Using Spatiotemporal Data with Machine Learning Models. Atmosphere, 14(10), 1517. https://doi.org/10.3390/atmos14101517




DOI: http://dx.doi.org/10.30970/eli.34.13

Refbacks

  • There are currently no refbacks.