Urbanization processes and the increasing number of vehicles in cities have made traffic management a major challenge for urban infrastructure. Traditional traffic light control methods, including fixed-time control, do not take into account the stochastic and dynamic characteristics of real-time traffic flows, resulting in delays and congestion. This paper explores an adaptive traffic control system by synthesizing the classical Webster model with modern machine learning algorithms. The research methodology involves the use of machine learning models to collect high-frequency traffic data from sensors and predict future traffic intensity. The prediction values are integrated into Webster formulas, allowing for dynamic determination and application of green signal duration in real time. The simulation modeling results show that the efficiency of the intersection is significantly improved, and the average vehicle delay and queue length are significantly reduced compared to traditional fixed-time control methods. The theoretical significance of the presented study lies in strengthening the connections between the two methods, namely traditional optimization and data-driven machine learning methods, and in practical terms, it is based on improving the efficiency of traffic flow management at intersections. By implementing this dual-approach system, city governments can move from reactive to proactive management, thereby increasing the stability and efficiency of the transport network.