Research On Highway Fare Evasion Vehicle Recognition Algorithm Based On Machine Learning

By the end of 2020,the total mileage of China’s highways reached 161,000 kilometers,with the continuous promotion of China’s network charging,highways have achieved the operation mode of "one network" nationwide,drivers only need to pass through the entrance and exit and entrance toll stations once,and the network toll collection fundamentally solves the problem of traffic congestion caused by manual tolling,greatly improves road traffic capacity,reduces energy consumption and environmental pollution caused by parking payment,but due to the impact of economic interests and the failure of public governance,Fee evasion of all kinds is endless.To a certain extent,fare evasion behavior destroys the normal traffic environment and even affects social stability,and the traffic data of highways contains a lot of information,and it is of great practical significance to study the identification of fare evasion vehicles according to the characteristics of data and actual problems.Based on the practicability of the classification model in machine learning,this paper constructs a high-speed vehicle fare evasion recognition model based on the practicability of the classification model in machine learning,and the specific work is as follows:Firstly,on the basis of relevant theoretical concepts,the traffic data of the 2020 desensitization in the research area is selected as the research data,the 38 feature fields in the original data are screened,the research fields required for this paper are preliminarily selected,and then the data is preprocessed to ensure the quality of the data input by the subsequent model,and then the characteristics of the fare evasion vehicles in the study area are analyzed,from the registration situation,license plate color,exit peak hour,road section,The 11,900 pretreated fare evasion vehicles were analyzed from the perspective of the amount of fare evasion,which prepared for the following identification model of fare evasion behavior using important features.Secondly,using the importance of random forest dimension to retain six features with high importance as independent variables of the identification model,according to the working principle of the classification model,five classification models of decision tree(DT),random forest(RF),gradient boosting tree(GBDT),BP neural network and extreme gradient ascent tree(XGBoost)are established in turn,and the results are compared by each evaluation index,in order to further study the recognition effect of fare evasion methods,this paper focuses on comparing the AUC values of each fare evasion method.It is found that the XGBoost model is significantly better than other models for the identification of most fare evasion methods,among which the AUC values of the three fare evasion methods of U/J type,Chong gang and large car and small standard are 0.9294,0.7561 and 0.9500,respectively,while the GBDT model has the best recognition effect on fake green pass fare evasion,and the corresponding AUC value is 0.8677.Finally,in order to further improve the accuracy and applicability of XGBoost model in the fare evasion method,based on the previous research,this chapter proposes a fare evasion method recognition model based on RFE-OPTUNA-XGBoost,on the one hand,the feature extraction method is replaced,and the recursive elimination algorithm(RFE)is used to select the features in order to find the optimal feature variables.On the other hand,the hyperparameter optimization of the XGBoost model is carried out by OPTUNA framework,and the accuracy of the optimized fare evasion method recognition model is as high as 0.945,and the average AUC values of each fare evasion method are 0.99744 for large car and small standard,0.98031 for U/J type,0.96893 for false green pass,and 0.92369 for chong gang.The results show that the RFE-OPTUNA-XGBoost model has a higher accuracy in identifying fare evasion methods and the AUC value of each evasion method.In summary,the identification model of highway fare evasion vehicles based on RFE-OPTUNA-XGBoost constructed in this paper can accurately identify the fare evasion method,which is of great practical significance for the highway management department to carry out audit work in practical applications.