Smart Traffic Light System Using Machine Learning

In Lebanon, traffic problems are a major concern for the population. The rising number of cars that exceeds the capacity of the roads, the inefficiency of public transportation infrastructures and the non-adaptive traffic light systems are contributors to the traffic crisis. Most roads in Lebanon suffer from traffic jams due to the traditional static green and red times allocations that are inconsiderate to the current state of the traffic. A solution to this problem is a system that adapts to the variations of the traffic dynamically and updates the traffic signal phases accordingly. In this paper, an adaptive traffic light system is implemented using reinforcement learning and tested using real data from Lebanese traffic. For training and testing the system, a software simulation tool is used. This tool can simulate the traffic intersection and allows the neural network to interact with it. Compared with the actual traffic light system, the proposed model displayed a reduction in average queue lengths by 62.82% and in average queuing time by 56.37%.