Effective parameter estimation and low computational complexity are the two major challenges involved in traffic light control. Most traffic light scheduling strategies focus on developing well-tuned off-line solutions. This paper focuses on the design of a hybrid traffic light control strategy. A macroscopic traffic network model is proposed to depict the traffic dynamics and a closed-loop traffic control strategy is designed based on the estimation of branching ratios at intersections. To reduce the computational complexity, a distributed algorithm is proposed based on the congestion level identification and system partitioning method, which is based on machine learning algorithms. Simulation results show the effectiveness of the proposed methodologies.