Vol. 5, Issue 1, Part A (2023)

This study investigates deep learning approaches for real-time traffic prediction in smart cities, addressing the growing need for accurate, time-sensitive forecasts to improve urban mobility. Using one year of real-time data from the City of Metronet’s traffic monitoring systems combined with open-source datasets, we developed and compared three models: Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and a hybrid Convolutional Neural Network-LSTM (CNN-LSTM). Data preprocessing included cleaning, feature engineering, and temporal encoding. Model performance was evaluated using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R², with statistical validation via paired t-tests and 5-fold cross-validation. The CNN-LSTM achieved the lowest RMSE (3.72) and highest R² (0.951), outperforming LSTM and GRU significantly (p<0.01). Analysis revealed CNN-LSTM’s ability to closely track both peak and off-peak traffic patterns, indicating robustness and adaptability. These findings highlight the potential of hybrid architectures in real-time smart city applications and support their integration into intelligent transportation systems to enhance traffic management and reduce congestion.