Enhanced Traffic Congestion Prediction Using Attention-Based Multi-Layer GRU Model with Feature Embedding

Intelligent Transportation Systems (ITS) are crucial for managing urban mobility and addressing traffic congestion, which poses significant challenges to modern cities. Traffic congestion leads to increased travel times, pollution, and fuel consumption, impacting both the environment and quality of life. Traditional traffic management solutions often fall short in predicting and adapting to dynamic traffic conditions. This study proposes an efficient deep learning (DL) model for predicting traffic congestion, utilizing the strengths of an attention-based multilayer Gated Recurrent Unit (GRU) network. The dataset used for this study includes 48,120 hourly vehicle counts across four junctions and additional weather data. Temporal and lagged features were engineered to capture daily and historical traffic trends and categorical data were considered by employing feature embedding. The attention-based GRU model integrates an attention mechanism to focus on relevant historical data, improving predictive performance by selectively emphasizing crucial time steps. This model architecture, consisting of two hidden layers and attention mechanisms, allows for nuanced traffic predictions by handling temporal dependencies and variations effectively. The performance was evaluated using various error metrics. The results demonstrate the model’s ability to predict traffic congestion with MSE of 0.9678, MAE of 0.4322, R² of 0.8686, MAPE of 6% offering valuable insights for traffic management and urban planning.

@article{M2025,
  title     = {Enhanced Traffic Congestion Prediction Using Attention-Based Multi-Layer GRU Model with Feature Embedding},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {16},
  number    = {1},
  year      = {2025},
  publisher = {The Science and Information Organization},
  author    = {Sreelekha M and Midhunchakkaravarthy Janarthanan},
  doi       = {10.14569/IJACSA.2025.0160194},
  url       = {https://doi.org/10.14569/IJACSA.2025.0160194}
}