Application Of Optimized Random Forest Algorithm In Farmers' Credit Risk Evaluation

The issue of "agriculture,rural areas and farmers" is the foundation of establishing a nation.It is the key to achieving the well-off goal.One of the important ways solving the "three rural issues" is to carry out the farmers credit.Since the 21 st century,the development of rural economy has accompanied the growing demand for credit in rural areas.It makes the credit of rural households gradually play a crucial role in supporting rural finance,boosting farmers' incomes,developing characteristic towns and reducing the gap between urban and rural areas.Farmers credit is essentially a kind of loan service.Under the situation that the rural risk control system is not perfect,the credit default situation of the farmer happens from time to time.How to effectively control and evaluate the credit risk of the farmer has become a top priority.In view of the credit problems of farmers,the current measures taken by rural financial institutions are not satisfactory.Based on this situation,the paper establishes a new mechanism of credit risk evaluation,which can optimize and improve the management system of credit financial institutions and provide solutions for the risk stability.At present,the technical methods used to assess the credit risk of farmers are not uncommon.Especially in rural areas,subjective judgments and statistical judgments of experts are often used,such as Logistic regression model and Probit regression model.With the introduction of artificial intelligence technology in the data era,BP neural networks,genetic algorithms,clustering algorithms and support vector machines have gradually emerged in recent years.The other one of these methods is random forest.At present,random forest method is mainly applied in the field of engineering such as target feature extraction,image recognition and signal processing.This paper uses this method in credit risk evaluation,and uses AdaBoost algorithm to strengthen random forest method.Finally,this paper builds an strong classifier model of AdaBoost-RandomForest,which can improve prediction accuracy.Based on credit risk evaluation theory,this paper describes the characteristics and classification of credit risk,the development history of rural finance and the BPneural network what is widely used in credit risk evaluation,and constructs AdaBoost-RandomForest theory model.Then this paper conducts an empirical test,analyzes the influencing factors of credit risk in A province,establishes a reasonable index system,uses the ROC curve and AUC value to empirically evaluate the survey data of B,and compares experiment results of AdaBoost-RandomForest with random forest and BP neural network.Finally,this paper gives empirical conclusions and prospects.The prediction accuracy of AdaBoost-RandomForest reaches 88.75%,which is 19.50% higher than that of single random forest model and 16.83% higher than BP neural network model.