Urban Traffic Prediction Using Two-Branch Attention Networks With Adversarial Domain Adaptation

The prediction of urban traffic forms an essential element in present-day intelligent transportation systems since it assists transportation agencies with improved traffic management and congestion control. Standard forecasting algorithms that monitor traffic experience difficulty in reaching accurate results because urban road networks show complex dynamic characteristics along with weather conditions and incidents on roads. Machine learning and deep learning models experience decreased accuracy when applied to different urban domains primarily due to domain shift problems. This study introduces Adversarial Domain Adaptation Two-Branch Attention Network (ADAT-BAN) as a new solution for urban traffic prediction to overcome existing challenges. The model achieves superior learning process performance through the implementation of the Starling Murmuration Optimization (SMO) algorithm. Optimized traffic parameter selection is achieved through the implementation of hybrid Tasmanian Devil Optimization (TDO) with Zebra Optimization Algorithm (ZOA) to select the most appropriate traffic metrics used for prediction purposes. The solution addresses domain shift problems through effective strategies to identify complex temporal and spatial traffic relationships. The framework achieves exceptional results through experimental testing because it produces prediction accuracy reaching 9 9. 9 \% above existing models. The framework demonstrates high operational capacity toward real-life traffic management in cities through improved urban flow management along with exact predicting capabilities.