Adaptive Traffic Signal Control System Using Machine Learning and Neural Evolution: Multi-Case Study for Intersection Optimization and Emergency Vehicle Prioritization

Urban traffic congestion poses persistent challenges, aggravated by insufficient integration of adaptive control and emergency vehicle prioritization. Existing AI models largely focus on either traffic flow optimization or emergency response, rarely combining both. This research fills this gap by developing a unified framework that predicts optimal green light durations using real-time traffic density, vehicle heterogeneity, and emergency prioritization. A four-way intersection traffic simulation was developed using the pygame library to closely mimic real traffic dynamics, lane configurations, vehicle types, and turning movements. Five case studies were conducted: (1) a mathematical green-time prediction model incorporating weighted vehicle classes, clearance times, arrival rates, and queue spillover controls; (2) reinforcement learning (RL) trained on varied traffic conditions; (3) RL enhanced with emergency vehicle priority; (4) application of the Neuroevolutionary of Augmenting Topologies (NEAT) algorithm for model architecture optimization; and (5) comparative analysis of model performance. The mathematical model reduced mean vehicle delay by 23% (p < 0.05), standard RL achieved 31% improvement (p < 0.01), emergency-aware RL maintained a 28% reduction while ensuring emergency vehicle clearance within 18.7 seconds on average, and the NEAT-based system improved throughput by 34% with superior adaptability to traffic fluctuations. The integrated framework significantly enhances traffic efficiency compared to fixed-time signals while guaranteeing rapid emergency vehicle response. Its extensibility supports integration into smart city traffic management platforms, offering scalable and adaptive solutions for urban mobility challenges.