DAT: Dual-Aspect Transformer for Travel Time Prediction

In the field of Intelligent Transportation Systems (ITS), reliable traffic prediction is crucial for enhancing planning and management strategies. Within this domain, significant research focuses on travel time prediction, utilizing deep learning techniques to make significant progress in forecasting for both short-term and long-term periods. Despite these advancements, there is still room for improvement to meet the dynamic requirements of ITS effectively. This study introduces the Dual-Aspect Transformer (DAT) for predicting travel times across short-term and long-term horizons in ITS. Leveraging transformer models' strengths, DAT effectively processes sequential data and captures long-range dependencies, offering significant improvements over existing methods. Extensive tests on a large real-world dataset demonstrate DAT's robustness, outperforming existing approaches in short-term forecasts and proving comparable to leading methods in long-term predictions.