Abstract
Traffic congestion in urban areas is a major challenge that leads to real-time problems such as elevated pollution level, greater fuel consumption, higher chances of vehicle collisions and stressed drivers. Therefore, Intelligent Transport Systems (ITS) are needed to improve the efficiency of transportation networks, for smoother and faster travel while reducing strain on existing infrastructure. Smart Traffic Light Systems (TLS) are a critical component of ITS, helping to achieve these objectives. Adaptive Traffic Light Systems (ATLS) employ an approach that dynamically adjusts signal timings based on real-time traffic demand. Compared to traditional TLS with fixed timing schemes, ATLS is capable of meeting the dynamic requirements of ITS. This paper proposes an Adaptive Traffic Light System that predicts traffic volume for the current hour and day using machine learning. It integrates this prediction with a pressure-based method that dynamically controls traffic light phases based on present traffic conditions. A simulation environment was developed using the Simulator of Urban Mobility (SUMO) to evaluate and validate the proposed approach at an isolated intersection with traffic demand and patterns varying by hour and day. To select the most suitable traffic prediction method, a comparative study was conducted among Random Forest, K-Nearest Neighbors, Decision Tree, Gradient Boosting, and XGBoost algorithms. The proposed system was compared with recent methods with similar objectives across 12 different scenarios. Results showed an average reduction of 26.3% in average waiting time, 22.4% in average time loss, 19.4% in total time loss, 23.8% in average CO emission, and 17.4% in average CO2 emission.