Research On Customs Risk Assessment Model Of Soybean Processing Trade Enterprises Based On Machine Learning Method

Since China's accession to the World Trade Organization,the volume of foreign trade has greatly increased,and the task of customs supervision has been aggravated.In actual supervision,the trade volume punished for violating laws and regulations only accounts for a small part,and indiscriminate inspection will cause waste of resources.In order to rationally allocate limited resources and improve the efficiency of supervision,the customs introduce risk management.Through enterprise customs risk assessment,the customs will focus on the trade of enterprises with potential risks and provide trade convenience for law-abiding enterprises.However,the existing customs risk assessment methods mostly use manual methods,which are inefficient,and the assessment results are easily affected by subjective factors.The flexibility of the established assessment model is poor.In addition,the existing customs risk assessment model has complex indicators,which depend on the internal data of enterprises and the historical data accumulated by customs,which is difficult to obtain in the assessment process and unable to evaluate newly registered enterprises without accumulated data.To solve the problems,this study takes soybean processing trade enterprises as an example,and looks for more accessible enterprise public information as evaluation indicators and uses machine learning method to build an evaluation model.A simpler and more efficient customs risk assessment method that does not depend on enterprise internal data and customs historical behavior data is formed,which makes the evaluation model results more open,reduces the difficulty of obtaining indicators and ensures the universality of evaluation for all enterprises.In this study,a new index system of enterprise customs risk assessment is constructed by means of literature investigation and expert consultation,taking into account the availability and quantification of indicators.According to the data of soybean processing trade enterprises,the random forest algorithm is used to further screen out eight indicators which have important influence on the evaluation results.In view of the imbalance of sample data,this paper uses SMOTE algorithm to resample samples,and trains the original samples and resampled samples by Naive Bayes,Adaboost and SVM respectively.According to the comparative analysis results,this paper determines the risk assessment model of soybean processing trade enterprises based on SMOTE + Adaboost model.Finally,the results of the risk assessment model of soybean processing trade enterprises are applied to the enterprise portrait,and the enterprise risks are displayed visually through the enterprise portrait so as to assist the customs supervisors in making decisions.The main contributions of this study are as follows:(1)The existing customs risk assessment methods are systematically combed,and the differences and shortcomings are analyzed and compared.The existing customs risk assessment mostly adopts analytic hierarchy process and fuzzy comprehensive evaluation method,which is cumbersome,subjective and inflexible;The machine learning method is efficient and convenient,and has a good performance in customs risk assessment.(2)A new index of enterprise customs risk assessment model is determined.The existing customs risk assessment model involves complex indicators which are difficult to obtain.This study synthesizes the existing research results and expert opinions,eliminates and adjusts the indicators,and finally extracts 26 indicators from 61 existing indicators.Based on the data,the random forest algorithm is used to further screen the indicators,and the redundant data indicators are eliminated.Finally,eight enterprise customs risk assessment indicators are determined which ensure the effectiveness and representativeness of the indicators.The formed indicators meet the requirements of accessibility and quantification,and are applicable to all enterprises.At the same time,the waste of resources caused by the customs in the process of obtaining enterprise indicators information and the cost of customs supervision are reduced.(3)The risk assessment results of enterprises are presented to the customs in the form of portraits,which is more intuitive.The risk model calculates the enterprise risk in the background,and outputs the final result to the front end in the form of enterprise portrait.Customs officers do not need to pay attention to complex enterprise data and the specific operation process of the model,but can clearly and intuitively obtain the enterprise information and risk level that they focus on by clearing the enterprise portrait.