A Spatial Skip-Connected Attention-based LSTM for urban traffic forecasting using air quality data
The rapid evolution of smart cities requires intelligent systems that are capable of addressing the growing complexity of urban environments. Urban traffic forecasting in smart cities remains challenging because of complex spatiotemporal dependencies and the presence of noisy, incomplete, and spatially heterogeneous sensor data. Existing forecasting models fail to preserve multiscale spatial relationships and long-term temporal dependencies, thereby reducing prediction accuracy under urban sensing conditions. To overcome these issues, this research proposes a Spatial Skip-Connected Attention-based Long Short-Term Memory (SSCA-LSTM) framework for traffic forecasting using air quality data. The proposed model integrates a U-shaped hierarchical spatiotemporal structure with spatial skip connections and spatial weighting mechanisms, thereby enabling it to learn both spatial and temporal dependencies at multiple urban scales. The spatial skip connections ensure gradient stability and fine-detail preservation at encoding and decoding, whereas the spatial weighting module adaptively identifies high-impacted regions based on spatial significance, thereby enhancing adaptive learning. Moreover, the hierarchical structure provides multiscale feature abstraction, capturing macro-level trends such as traffic flow and micro-level variation such as pollution patterns. Experimental analysis of the CityPulse EU FP7 dataset demonstrates that the proposed SSCA-LSTM achieves a Mean Square Error (MSE) of 0.3310 and a Coefficient of Determination (R²) of 0.927, outperforming existing models.
Authors
- Miguel García-Torres (ORCID: https://orcid.org/0000-0002-6867-7080)
- Sanjay Shekar Nagenahalli Chandrashekar
- Vishwanath Bidare Ranganath
Institutions
- Visvesvaraya Technological University (IN)
- Universidad Pablo de Olavide (ES)
Publication Details
- Journal
- Discover Computing
- Published
- 2026-11-21
- DOI
- https://doi.org/10.1007/s10791-026-10812-8
- Primary Topic
- Traffic Prediction and Management Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00