The Research On Farmer Household Credit Evaluation Based On Machine Learning

In recent years,under the guidance of the policy of building a moderately prosperous society in all respects,the state has been making more and more targeted efforts to alleviate poverty,and the lives of poor farmers have changed,however,the key to the problem of "agriculture,rural areas and farmers" is the issue of agriculture,for the broad masses of farmers,development of agricultural economy is still rich in the main roads,and the capital is the core elements of the agricultural economy development,farmer’s source of funds is relatively limited,through financial credit institutions such as bank loans remains the main channel to get funding,they and the rural financial system is relatively weak,subsequently led to a "double difficult" problem,namely "hard loan" rural financial credit institutions and farmers "loans difficult".The main reasons behind this are the information asymmetry between farmers and financial credit institutions,the imperfect rural credit assessment system,the inability of rural financial credit institutions to accurately and objectively assess the credit of farmers,and the high default rate of farmers’ loans.Therefore,under this background,it is of great significance to explore a set of scientific and unified credit evaluation system applicable to Chinese farmers to solve the problem of "double difficulty" and solve the problem of "agriculture,rural areas and farmers".Specifically,this paper expounds the research background of farmer credit evaluation,and respectively from the Angle of theory and practice,the significance of the study of farmer credit evaluation,then the credit evaluation,credit of farmers,farmers credit evaluation indexes of farmer credit evaluation method and the selection of relevant literature to comb,and the solid foundation for further study.Secondly,the theory of machine learning is systematically expounded.Including the concept and development of machine learning,the theoretical basis of the machine learning algorithm used in this paper,the parameter adjustment method used in machine learning and the evaluation index involved in the machine learning algorithm.Again,the concepts of farmer households,farmer households credit and farmer households credit risk are explained,and according to the literature of the existing research results,from the basic characteristics of farm households,the repayment capacity of farm households,and the guarantee of farm households,and the stability of the four dimensions of farm households constructing farmer credit evaluation system,and according to the relevant standards to meet the requirements of this article research evaluation indexes for a total of 18.Then to Chinese family financial investigation and the basis of the research center of peasant household survey data,collected from farmers large data sets,this paper research on the preprocessing data sets(including the missing value treatment,abnormal value,consistency processing)and exploratory analysis(including target and characteristic variables).The preparing work before then,modeling derivatives(including structure variables,standardization of data sets and discretization processing,One-Hot Hot coding processing and divided into training set and testing set),and then build the farmer credit evaluation model based on machine learning,this article mainly USES the Logistic regression,decision tree,the Random forest as well as GBDT four machine learning algorithms model of concrete is through the confusion matrix of binary classification prediction,the credit status of the farmers at the same time through the regularization,the grid search and cross validation method tuning parameter optimization for each model to adjust,The evaluation indexes AUC,KS,PSI,recall,precision and F1 were used to conduct a comprehensive comparative analysis of the models.Finally,the farmers’ credit score card based on Logistic regression was introduced to quantitatively score and classify the farmers’ credit.Empirical research shows that:(1)the evaluation effect of each model has been significantly improved after parameter adjustment,indicating that regularization,cross validation and grid search and other parameter adjustment methods have positive significance to the improvement of the model and are worth promoting.(2)the machine learning model showed good effects on the scores of evaluation indexes AUC,KS,PSI and F1 with a small gap,indicating that the machine learning model has a good prediction ability in the credit evaluation of farmers,and has a certain application prospect compared with the traditional expert experience judgment method which is complex,time-consuming and inefficient.As far as comprehensive effect is concerned,the model of integrated classification algorithm is relatively superior while the model of decision tree is relatively poor.Among them,the Logistic regression model had the best performance on recall and the integrated classification algorithm model had a better performance on Precision.(3)farmers’ credit is divided into four grades from low to high: grade D farmers should refuse their loans due to high default risk,grade C farmers should be cautious loans due to high default risk,grade B farmers’ default risk should be further examined before deciding whether to issue loans,and grade A farmers with very low default risk can issue loans.Finally,based on the research,this paper puts forward the research prospect and puts forward the following policy Suggestions for farmers,financial credit institutions and government departments.Second,financial credit institutions should establish a unified credit evaluation standard and introduce a new machine learning method to farmers’ credit evaluation.Third,the government should establish a mechanism for farmers’ credit information sharing and transmission.Fourth,the government also needs to work with financial credit institutions to establish a credit reward and punishment mechanism for farmers,and strive to promote the construction of the credit environment.