In order to address the problems of low efficiency and high cost in the existing highway toll inspection, a K-nearest neighbor (KNN) -based recognition model of toll evasion events was proposed. Firstly, the characteristics of toll evasion behavior of large vehicles but small signs on highways were analyzed, the recognition and verification algorithms were designed and established to extract toll evasion samples from the original toll data. Secondly, after preprocessing the original evasion data, principal component analysis was used for feature dimensionality reduction. Finally, in view of the extreme imbalance of sample size distribution in different categories of the evasion dataset, the borderline2 synthetic minority over-sampling technique algorithm (BorderlineSMOTE2) in the oversampling method was used to balance the data, and the KNN algorithm was used to establish a classification and recognition model for evasion behavior. The final verification results show that the recognition accuracy rate of the established toll evasion behavior recognition model is 0.75, the recall rate is 0.84, and the f₁ -score is 0.79, indicating that the model has a higher classification and recognition accuracy for the toll evasion behavior samples and better model performance. The recognition model for highway vehicle evasion incidents based on KNN has established corresponding processing rules and algorithms to address the high-dimensional imbalance characteristics of evasion data, improving recognition accuracy. The recognition results can assist highway toll inspection in effectively screening evasion behavior and reducing the cost of toll loss.