Abstract: To solve the problems of high cost and generalization feasibility existed generally in current cargo vehicle mass calculation methods, an innovative method was proposed to estimate the mass of heavy trucks. Integrating vehicle dynamics theory and machine learning algorithms, the method was arranged to train and validate the model based on the supervised learning methods and high-speed traffic big data. Firstly, a cluster analysis was conducted to determine the thresholds for distinguishing among unloaded, partially loaded, and fully loaded vehicles, providing an important basis for subsequent calculations. Then the random forest algorithm was used to train the constructed classification model to determine the basic loading status of vehicles during a given trip, and select the stable driving segments from the vehicle driving data, calculating the vehicle mass of these segments according to vehicle system dynamics theory. Finally, the segments were filtered and aggregated based on the loading status classification results to obtain the final vehicle mass calculation. The study results show that, validating high-speed traffic big data, the proposed method can control the mean absolute percent error(MAPE) of trip mass calculations within 10% for both unloaded and fully loaded states, exhibiting high accuracy. Compared with the existing technologies, the method does not require the installation of additional sensors and the requirements for data collection, storage, and computing equipment are relatively low, offering significant cost advantages. Therefore, the method developed based on big data can provide a potential for rapid and widespread adoption in traffic regulation, logistics transportation, and product development.