A new deep learning framework called Spatial Skip-Connected Attention-based LSTM uses air quality measurements to forecast traffic congestion in cities. The model, tested on the CityPulse EU FP7 dataset, achieved a coefficient of determination of 0.927 and an MSE of 0.3310, outperforming traditional recurrent and graph-based networks. By linking pollutant levels like carbon monoxide and nitrogen dioxide to vehicle activity, it offers a low-cost way to monitor traffic without relying solely on dedicated sensors. The approach could help city planners and emissions managers anticipate congestion and adjust traffic flow or policies, potentially reducing fuel waste and pollution. The model also maintains accuracy across forecasting horizons up to 24 hours and remains stable even with 30% missing sensor data. While the paper focuses on technical performance, it opens the door for practical applications in smart city environmental monitoring and emissions reduction strategies.
