Bidirectional Temporal Attention Networks for Missing Data Recovery in Urban Air Quality Monitoring
INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT · 25.05.2026 · 10 (05) · 1-9
DOI: 10.55041/ijsrem63739 ↗Kurz gefasst
Fachartikel in der Zeitschrift INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT, online erschienen am 25.05.2026. Verfasst von Aklesh Shadaf und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Urban air quality sensor networks produce continuous multi-pollutant time series that are routinely interrupted by hardware failures, power outages, and scheduled maintenance, creating missing data gaps that compromise downstream analysis, regulatory reporting, and health exposure assessment. Existing imputation approaches — from classical statistical interpolation to recurrent neural networks and variational autoencoders — predominantly exploit only past observations and fail to leverage future context that is available at imputation time. In this paper we propose BiTAN, a Bidirectional Temporal Attention Network that applies stacked self-attention encoders independently in the forward and backward temporal directions and fuses their representations through a learned cross-directional layer. BiTAN incorporates variable-type-specific learnable missing embeddings, motivated by recent variational approaches [1], enabling the model to represent the absence of each pollutant or meteorological variable distinctively. Evaluation on EPA Air Quality System data from 12 Chicago-area monitoring stations under multiple missing-data regimes demonstrates 17–21% RMSE reduction over recurrent baselines, 17.9% reduction over the variational baseline [1], and competitive computational efficiency. Ablation experiments confirm the contribution of bidirectionality, the fusion layer, learnable embeddings, and positional encoding individually.
Zitieren
APA
Shadaf, A., Paurja, S., & Singh, A. (2026). Bidirectional Temporal Attention Networks for Missing Data Recovery in Urban Air Quality Monitoring. INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT, 10(05), 1-9. https://doi.org/10.55041/ijsrem63739
Vancouver
Shadaf A, Paurja S, Singh A. Bidirectional Temporal Attention Networks for Missing Data Recovery in Urban Air Quality Monitoring. INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT. 2026;10(05):1-9. doi:10.55041/ijsrem63739
BibTeX
@article{Shadaf2026,
author = {Shadaf, Aklesh and Paurja, Sandeep and Singh, Ashish},
title = {Bidirectional Temporal Attention Networks for Missing Data Recovery in Urban Air Quality Monitoring},
journal = {INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT},
year = {2026},
volume = {10},
number = {05},
pages = {1--9},
doi = {10.55041/ijsrem63739}
}
EUVIMED