Research On Credit Evaluation Of Family Farms And Ranches Under Complex Information Fusion

By the end of 2022,nearly 4 million family farms had been registered in the national family Farm Directory system,and more than 170,000 demonstration family farms had been established at or above the county level.Family farms,as an important part of new agricultural operations,have become the main force promoting the development of modern agriculture and an indispensable force in the process of agricultural modernization.Compared with the agriculture-oriented family farms in other provinces,the Inner Mongolia Autonomous Region shows the development characteristics of family farms and ranches(FF&R)that combine agriculture and animal husbandry with planting and breeding.Therefore,family farms in the Inner Mongolia Autonomous Region are directly reflected as FF&R.In the process of continuous development of FF&R,financing difficulties are always the main problem that troubles the sustainable development of FF&R.The development of FF&R cannot be separated from the support of financial system,and the construction of scientific and reasonable credit evaluation mechanism plays a fundamental role in solving this problem.Firstly,by analyzing the development status,credit status and construction of credit evaluation system of FF&R in Inner Mongolia,this thesis explores the formation mechanism of credit risk of FF&R in Inner Mongolia,expounds the information sources and information collection standards of FF&R in Inner Mongolia,and then constructs a credit evaluation index system of FF&R with default discrimination ability.This thesis uses classical comprehensive evaluation method and machine learning evaluation method to evaluate the credit of FF&R in Inner Mongolia.Finally,the stability of different evaluation methods is determined based on the "horizontal and vertical perspective",and the traditional credit evaluation methods are combined with the popular machine learning evaluation methods through information aggregation,so as to integrate the advantages of various methods to improve the reliability and credibility of evaluation results.The main work of this thesis is as follows:(1)Construct the credit evaluation index system of family farm and pasture with both subjective and objective perspectives.Firstly,32 credit evaluation indicators of FF&R were initially screened based on the financial characteristics and management characteristics of FF&R,combined with the five principles of high frequency indicators and the comprehensiveness and operability of index screening in previous references.Secondly,the depth weighted fuzzy Bayesian model is used to conduct secondary screening on the indicators of the preliminary screening,and select the credit evaluation indicators that can significantly distinguish the multi-classification default status(non-default,low default,high default)of FF&R.On the one hand,the constructed credit evaluation index system of FF&R can fully reflect the financial characteristics of FF&R;on the other hand,through rigorous data analysis,a credit evaluation index system of FF&R with default discrimination ability can be obtained.(2)The introduction of traditional credit evaluation methods and machine learning evaluation methods for FF&R credit evaluation.For a long time,there has been a lack of effective credit evaluation and supervision for FF&R,so the judgment of credit status of FF&R mostly relies on expert experience evaluation and subjective judgment of business personnel.Due to the lack of development of the current credit evaluation methods of FF&R,the data of FF&R are prone to distortion and fuzziness,etc.This thesis firstly introduced the entropy weight TOPSIS evaluation,DEA evaluation and other classical evaluation methods by referring to the commonly used credit evaluation methods of commercial banks,financial institutions,small and micro enterprises,etc.Construct FF&R credit evaluation model;The second is to introduce the current popular machine learning credit evaluation methods such as support vector machine and neural network to build the FF&R credit evaluation model,and study the application of the credit evaluation model based on data mining in the FF&R credit evaluation.(3)Various evaluation methods of different types are analyzed in detail from the perspective of stability.This thesis analyzes the stability level of selected classical credit evaluation methods and machine learning evaluation methods,reveals the strength of the stability of different credit evaluation methods,and finds the most suitable method for FF&R credit evaluation from the perspective of stability.According to STDEV.A stability index,the stability ranking of credit evaluation methods from the horizontal perspective is as follows: neural network method > support vector machine method > entropy weight-TOPSIS method >DEA method.From the longitudinal perspective,the stability order of credit evaluation methods is as follows: entropy weight TOPSIS method > neural network method > support vector machine method >DEA method.Therefore,it can be determined that compared with the classical credit evaluation methods,the machine learning model has better stability in the application of FF&R credit evaluation from the horizontal and vertical perspectives,and DEA model is general in the application of FF&R credit evaluation.(4)A combination of evaluation methods for FF&R under complex information fusion was constructed.In the existing evaluation field,due to the difference of the action mechanism and information selection of different evaluation methods,the evaluation results of various methods are inconsistent.Although this thesis analyzes and judges the advantages and disadvantages of the stability of different credit evaluation methods based on the perspective of stability,no matter which method we choose,it may cause potential information loss.Therefore,it is necessary to combine different credit evaluation methods to avoid the problem of inaccurate information caused by the use of a single method.In this thesis,by combining various evaluation methods of different mechanisms of classical credit evaluation and machine learning evaluation,the mean value method is used to construct a combination evaluation optimization model to aggregate the similar characteristics of various evaluation methods,integrate the advantages of various methods and solve the problem of inconsistent conclusions of various evaluation methods.