Application Of Ensemble Learning Algorithm In Personal Credit Assessment

With the development of the economy,most people use credit cards,ants borrow and other new products to consume in advance so as to change the traditional consumption concept of "living within one's means".At the same time,a variety of loan methods have begun popular,and various types of online and offine funds lending have become more and more common.However,from another perspective,the prevalence of a situation often causes a series of new problems.Nowadays,the popularity of credit consumption is no exception.The resulting customer default has seriously affected the development of the credit industry,which has caused certain obstacles to the development of the national economy.The occurrence of this phenomenon results in the existing credit evaluation system face enormous challenges,and the issue of personal credit assessment has gradually become a research hotspot in the credit industry.Therefore,the introduction of more effective assessment methods will contribute to economic development.The customers who apply for loans are mainly divided into two categories,namely,default and non-default.We can deal with that as a classification problem.The main purpose of this study is to establish a classification model based on the personal economic conditions,property conditions and historical credit records of the loan applicants,so that we can obtain the evaluation of customer credit.This paper systematically preprocesses the set of financial credit data provided by a foreign bank and a domestic lending institution.Bayesian optimization algorithm is used to optimize the parameters of the model.In addition,three different models are proposed to analyze the importance of the features.Then,we train the model based on three representative learning algorithms(there are Bagging,Boosting and Stacking).Among them,Stacking uses random forest,support vector machine and LightGBM as the primary learner,and logical regression as the secondary learner to build the evaluation model.Besides,some single classification algorithms are also used to compare the performance with ensemble learning methods,including logistic regression,support vector machine,k-nearest neighbor,decision tree and neural network.Finally,the performance of these models are compared by using the Friedman and Nemenyi testing methods,and further verified depend on the Friedman test chart.Based on the above model,we use a ten-fold cross-validation method to verify performance characteristics.The results show that the models trained by the Bagging and Boosting ensemble learning algorithms have played a certain role in improving the classification performance of the system,and the latter has a better performance.However,the effect of Stacking ensembling multiple models does not achieve the expected results,and fails to effectively improve the classification performance.All in all,the ensemble algorithm is better than the single in terms of classification performance relatively.According to the comparison results of the two test methods,it is found that the performance of some algorithms is different under the condition of a confidence of 95%.