Traffic Management System Using Machine Learning and Ant Colony Optimization

Wong, Sheng Wei (2025) Traffic Management System Using Machine Learning and Ant Colony Optimization. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text WONG SHENG WEI_FULL TEXT.pdf Restricted to Registered users only Download (2MB)

Abstract

In order to overcome the high computational cost and latency of metaheuristic traffic-signal timing, this thesis presents a two-stage framework that couples an Ant Colony Optimization (ACO) algorithm developed and tuned in Jupyter Lab with a Random Forest Model (RFM) trained in the same Jupyter Lab. ACO was used to generate optimized signal timing for a prototypical fourway intersection under varying queue-length and weather conditions. ACO requires on average 11.2s run time, while the RFM predicts equivalent timings in just 373.4 ms—achieving a 96.7 % reduction in run time. When evaluated in SUMO simulations across normal, light-rain, and heavy-rain scenarios, the signal timing predicted by RFM cleared congestion within 5 % of the ACO benchmark (515s vs 579s, 770s vs 862s, and 1028s vs 1067s, respectively). These results demonstrate that a properly trained machine-learning model can approximate metaheuristic outputs with negligible loss of control quality while enabling sub-second updates, paving the way for scalable, real-time adaptive signal control in smart-city environments. Limitations include reliance on simulated data for a single intersection, the need to retrain for different layouts, and sensitivity of RFM predictions between deviations of 2s to 3s, which may affect performance under extreme traffic congestion.