Urban traffic congestion remains a major challenge in intelligent transportation systems, requiring adaptive and data-driven traffic control strategies. This study proposes a simulation-based adaptive traffic signal control framework using deep learning and metaheuristic optimization. The framework integrates YOLOv8-based vehicle detection with a Kernel Extreme Learning Machine classifier optimized using Particle Swarm Optimization, Genetic Algorithm, Artificial Bee Colony, and Grey Wolf Optimizer. Vehicle counts and classes extracted from traffic images determine adaptive signal timings within a custom simulation environment. All optimization algorithms were evaluated over 30 independent runs using different random seeds. The Genetic Algorithm-optimized model achieved the highest classification accuracy (91.733%), and statistical analyses confirmed its superiority over competing methods. The YOLOv8 model achieved a mAP@50 of 92%. Simulation results demonstrated dynamic signal-time reallocation according to changing traffic demand compared with fixed-time control. The findings indicate that the proposed framework can support adaptive traffic management using existing roadside camera infrastructure.
GÖKCAN et al. (Sun,) studied this question.