Design and Evaluation of an AI-Driven Decision Support Architecture for Adaptive Traffic Signal Control in Urban Intelligent Transportation Networks

• Usman Javed

  Mehran University of Engineering and Technology, Department of Electrical Engineering, Jamshoro Road, Jamshoro, Pakistan

  Author

• Bilal Akhtar

  Sir Syed University of Engineering and Technology, Department of Electronic and Telecommunication Engineering, University Road, Gulshan-e-Iqbal, Karachi, Sindh, Pakistan

  Author

• Noman Siddiqi

  Balochistan University of Information Technology, Engineering and Management Sciences, Faculty of Telecommunication and Networking, Airport Road, Quetta, Balochistan, Pakistan

  Author

• Ariful Islam

  Dhaka, Bangladesh

  Author

Urban mobility increasingly depends on signalized intersections that must operate under variable demand, partial observability, and infrastructure constraints. Traditional timing plans, while reliable, are often insensitive to short-term fluctuations that shape travel time, queue length, and emissions. Recent advances in sensing, embedded computing, and learning algorithms suggest a path to decision support systems that adapt signal policies in real time. This paper presents a design and evaluation of an AI-driven decision support architecture for adaptive traffic signal control spanning perception, prediction, and control layers. The architecture integrates multi-source data via probabilistic state estimation, generates short-horizon demand forecasts, and selects phase and split decisions through an optimization-guided policy that balances throughput, delay, fairness, and actuation wear. The system emphasizes auditability and safety through constraint handling, interpretable surrogate models, and runtime monitors that prevent spillback and enforce clearance times. A learning component refines policy parameters with feedback from observed outcomes while preserving stability through conservative updates. The evaluation considers heterogeneous networks, diverse demand regimes, and exogenous disturbances, with a focus on congestion onset, recovery, and rare events. Results indicate consistent reductions in delay and queue variance without aggressive saturation, along with improved robustness to sensing dropouts and demand shocks. The discussion underscores trade-offs between exploration, safety, and interpretability, and provides guidance on configuration, calibration, and fail-safe operation. The study aims to inform neutral, pragmatic deployments where adaptive benefits are realized while maintaining predictable behavior under uncertainty.

• PDF

Copyright (c) 2025 authors

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.