Hybrid Machine Learning-driven Digital Twin Framework for Real-Time Traffic Optimization in Mega Cities

Abstract

Traffic congestion remains a major challenge in Lagos, causing substantial delays, higher fuel consumption, pollution and reduced productivity. As the city continues to grow and vehicle ownership increases, the existing road infrastructure is insufficient to accommodate rapidly rising transport demand. Intelligent traffic management solutions are therefore required. This study proposes a hybrid machine learning-driven digital twin framework for Lagos. The framework integrates digital twin technology, hybrid machine learning and reinforcement learning to improve traffic prediction, reduce congestion and optimise traffic flow. It processes traffic data generated from simulated sensor, camera and connected-vehicle environments representing Lagos traffic conditions. The model combines temporal learning for identifying time-dependent traffic patterns with spatial learning for capturing road-network relationships. A reinforcement learning module is also integrated to optimise traffic-signal timing and routing decisions dynamically. The framework focuses on major transport corridors, including Third Mainland Bridge, Ikorodu Road and the Ozumba Mbadiwe corridor, where recurrent congestion occurs. Testing was conducted using the Simulation of Urban Mobility platform and Python-based tools. The system was evaluated under representative Lagos traffic scenarios, including peak-hour congestion, traffic incidents and routine weekday operations. The experimental results showed improved performance compared with traditional approaches and standalone machine learning models. The proposed framework reduced traffic prediction error by 61.6% compared with baseline approaches. Congestion decreased by up to 40.3% during peak periods, while waiting time was reduced by 49.8%. Digital and physical traffic states remained synchronised at 96.8% within real-time processing requirements.