Research On Personal Credit Evaluation Based On Decision Tree Integration Algorithm

Together with the rapid development of the market economy,consumer credit is constantly growing;there is a need to create a universal financial infrastructure that covers all markets.In connection with the development of industry and trade,the banking system of the society,which carries out its operations in the banking market(the most important is the trade in credit resources)also gets development.It allows users to use various types of loans.In this case banks often face with a credit risk factor,when a person for various reasons is not able to repay his loan on time.Modern intelligent systems,based on machine learning,are able to provide banks with an opportunity for a better and more comprehensive assessment of the personal credit risk of customers.To make a decision,the bank needs a comprehensive and scientific understanding basic information about the customer and his credit history,as well as assessing the situation with loans,to control the risk,avoid possible debt and reduce the resulting losses.Currently,commercial banks use a credit score card to assess the creditworthiness of potential borrowers,which includes a number of general as and special issues,on the basis of which the report is generated on the client's creditworthiness.The purpose of this thesis is to explore the possibility of improving the existing methods for determining the assessment of the personal creditworthiness of potential borrowers based on classification algorithms for further use in the field of personal credit banking systems.When solving the problem,general methods of data mining,such as the classical decision tree C 4.5,which is the basis of this study,are considered.To identify potential borrowers and their probabilistic classification,existing machine learning approaches used to estimate a personal loan based on the probabilistic classifier Naive Bayes.For accurate distribution of potential borrowers,the algorithm of the single-dependent average AODE estimate is used.Due to which,according to the result of machine training,it is possible to accurately screen out potential borrowers into four main categories,helping to determine the exact number of clients who fall into the "risk zone".In result of the boxes giving the potential positives,the solubility is shown,and the status of the "positive" abo "negative" positives is revealed for the subordinate distribution on the "plateaus".A simple model of a tree is present for identifying the status of potential positions,recognitions for the banking system in the sector of credit.The developed model investigated by the software application Weka,which allowed visualizing the results of identification of potential borrowers.In the framework of research work on a simple model,a decision tree proposed,based on which the author constructed further studies.The use of existing methods of ranking significant attributes of classification models can significantly reduce the dimensionality of the input parameter vector of the model with virtually no loss of model accuracy,which reduces the costs of constructing a classification tree and the probability of a table in the process of machine learning.Separation of classifiers based on a decision tree and probabilistic classifiers makes it possible to build a more accurate classification model that allows reducing the share of these clients,in which each of the models separately cannot give an unambiguous decision on the possibility of lending.Practical value shows the possibility of reducing the dimension of the classification model without reducing its accuracy,which reduces the costs of constructing a classification tree and a probability table in the process of machine learning.