With the rapid expansion of the highway network, the phenomenon of toll evasion has become increasingly serious, leading to huge economic losses. This study constructs an efficient model for recognizing evasion vehicles by integrating random forest algorithms, logistic regression, and neural networks to improve inspection efficiency and reduce losses. This study first uses the Min-Max standardization method to process the data to eliminate the influence of different dimensions and numerical ranges. Secondly, key features are selected using the random forest algorithm, and the probability of evasion is predicted using a logistic regression model. Finally, the types of evasion are identified using neural networks. The integrated model performed better than other models in prediction accuracy, classification accuracy, inspection time, mean square error, root mean square error, and stability, as indicated by the results. By achieving predictions accuracy of 92%, recall rate of 94.71%, and AUC of receiver operation characteristic of 96%, it demonstrated excellent prediction performance and application value. Highway management departments receive reliable technical support from this study, which effectively improves the accuracy and efficiency of toll evasion inspections, helping to reduce economic losses and maintain traffic order.