Research On Personal Credit Evaluation Model Of Online Loans Based On LightGBM And HWBOA?BP

With the in-depth integration of the Internet and financial services and the upgrading of the national consumption structure,the advanced consumption model based on credit consumption has gradually become the mainstream way of people's daily consumption.The booming credit economy not only promoted the rapid growth of household consumption,but also gave birth to a large number of loan default accounts.In view of the huge risks that increasingly prominent loan defaults have brought to the consumer finance industry,scientific and objective assessment of the credit status of borrowers,whether it is to control the loan risk of consumer financial institutions or is positive for the credit economy's healthy development.In the actual online loan business,due to the extremely small proportion of loan default accounts in the total loan accounts,most of the current research on online loan credit evaluation models focuses on how to improve the classification accuracy of the evaluation model,ignoring the impact of the difference in feature importance on the accuracy of the credit evaluation model's classification.In addition,the online loan credit evaluation model also needs to have strong interpretability to help both borrowers and lenders understand the loan rules.Therefore,in view of the general problems in the current research on online loan credit evaluation,this paper starts research from two levels of balancing data and improving classification algorithms,and then builds an online lending personal credit evaluation model.The main content of the algorithm improvement level includes the following two points:1.A hybrid whale-bat optimization algorithm(HWBOA)based on adaptive strategy is proposed.In the problem of online loan credit evaluation,different parameter configurations have a certain impact on the classification accuracy of the evaluation model.When the traditional grid search algorithm is faced with various possible combinations of a large number of parameters,the time cost for finding the best parameter combination increases almost exponentially,and it is difficult to adapt to the needs of practical applications.Based on the above problems,this paper proposes the HWBOA algorithm by optimizing and improving the basic WOA algorithm,and test its optimization performance.The test results show that the improved algorithm has greater advantages in parameter optimization.2.An integrated classification algorithm(LHBP)based on Light GBM and HWBOA?BP is proposed.Net loan against an existing credit evaluation model can not take into account the problem of high classification accuracy and strong explanatory.Based on the BP neural network's characteristics of high classification accuracy,poor interpretability,and slow convergence speed,BP neural network is used as the basic model of classification.Light GBM and HWBOA are optimized and improved on BP neural network,and the LHBP ensemble classification algorithm is proposed,then uses the public data set is experimentally verified.Experimental results show that the LHBP ensemble classification algorithm combines the advantages of the above three algorithms,not only can screen out important features in a short time,and has strong interpretability;it also has a high classification accuracy and stability.The research content at the data processing level mainly includes the following two points:1.A SMOTE oversampling algorithm(SMOTE?HWBOA)optimized based on the HWBOA algorithm is proposed.In online loan credit evaluation,there is usually the problem of imbalanced loan samples.When using the SMOTE oversampling algorithm to balance loan data,subjective setting of the sampling ratio and the number of adjacent samples can easily blur the original classification boundary,resulting in the quality of the generated data is poor,which in turn reduces the classification accuracy of the credit evaluation model.Based on this,this paper proposes the SMOTE?HWBOA oversampling algorithm,and uses the public data set for experimental verification.Experimental results show that the quality of data generated by the SMOTE?HWBOA oversampling algorithm is better.2.The WCGAN-GP oversampling algorithm is applied to the problem of online loan credit evaluation.In view of the fact that traditional oversampling algorithms only use the local characteristics of minority samples when balancing online loan data sets,the resulting minority samples have a large amount of redundancy,which can easily cause overfitting of the credit evaluation model,and makes it impossible to objectively evaluate customer credit.Based on the characteristics of the GAN model that can generate approximate real data,this paper introduces the WCGAN-GP oversampling algorithm,which uses the trained and stable WGAN-GP as the benchmark model,and introduces restrictions on the basis to guide the generation of minority samples to further improve the quality of the generated data.The results of comparative experiments combined with public data sets show that compared with the traditional oversampling algorithm,the WCGANGP oversampling algorithm is more effective in solving the problem of sample imbalance.At the end of this paper,an empirical analysis of the online loan credit evaluation model is carried out.The source data of the empirical analysis comes from the "lending club" website,which is preprocessed first,and then the two oversampling algorithms proposed in this article are used to balance the data set.Secondly,an online loan credit evaluation model based on the LHBP integrated classification algorithm is constructed.Finally,the online loan credit evaluation model constructed in this paper is compared with the commonly used online loan credit evaluation models such as GBDT and XGboost.The experimental results show that after the same oversampling technology is processed,the online loan credit evaluation model proposed in this paper not only has better classification accuracy,but also has stronger interpretability;At the same time,it is found that compared to the SMOTE?HWBOA oversampling algorithm,the WCGAN-GP oversampling method is more suitable for the balance problem of online loan credit data sets.