Lane change behavior is one of the most foundational driving behaviors in microscopic traffic flow. Researching the lane change behavior contributes to improving the simulation accuracy of lane change models and reducing traffic accidents caused by improper lane change behavior. The current lane change model is the decision model mostly based on the way of driver's thinking. The shortcoming of current models is difficult to catch certain potential decision-making model and influencing factors in the driver's decision-making process. In view of this, this paper introduces a typical artificial intelligence method, Bayesian networks, to establish a new lane change model, and tries to improve the accuracy of the lane change model by machine learning. It uses a segmented discrete method to preprocess vehicle trajectory measurement data, and uses the processed data to training and verification this model. The verification results show that, this model's recognition rate to lane change behavior can reach more than 88%. In addition, this model can be further applied to the development of a driver assistance system.