In order to improve the efficiency of identifying toll evasion behavior by changing paths on highways, this paper conducts research on toll evasion behavior by changing paths. A model for identifying toll evasion behavior by changing paths on highways based on random forests is established, which can effectively identify such behavior of toll evasion and assist relevant management departments of highways in recovering evaded fees. Firstly, the original toll data are analyzed to filter out the fields related to this study, and the 12 initial features that can be inputted into the model are obtained after arithmetic; Secondly, the features with covariance are eliminated by calculating the Variance inflation factor (VIF) and Tolerance (TOL) values of each feature, and the Boruta algorithm is used to filter out the high-importance features ("whether the driving direction is consistent", "whether the entry and exit stations are consistent", "travel time", and "minimum fare mileage"); Thirdly, the data set is balanced using the SMOTETomek integrated sampling technique; Then, the grid search method is used to tune the hyperparameters of the random forest; Finally, the model built in this paper is utilized for training and recognition, and the recognition effect is compared with that of the benchmark model. The results show that the model developed in this paper can better recognize the toll evasion behavior by changing paths on highways, and the Macro-F1 score reaches 0.966, which is better than the extreme gradient boost (XGBoost) (0.9431), decision tree(DT) (0.9563) and gradient boosting decision trees (GBDT) (0.9382), and it can provide reference for operation management departments to inspect such toll evasion vehicles.