Modern traffic control strategies require knowledge of the vehicles’ density. However, when such data is available through sensors or cameras, it often lacks accuracy and completeness. In this context, we propose an enhanced cellular automaton that can provide labeled data in accordance with the three-phase traffic flow theory. This study leverages such model to address traffic state classification, which is valuable for adaptive traffic control. Specifically, the effectiveness of the graph neural network in using three-phase labeled data for traffic classification will be demonstrated by achieving high accuracy. This ensures a clear distinction between traffic phases and paves the way for further research on the factors affecting the traffic cellular automaton model1.