Research On Default Risk Early Warning Of Online Loan Platform Based On Machine Learning Algorithm

In recent years,the risks accumulated by online loan platforms have begun to appear.Many normal operating online loan platforms are not strictly audited,often rely on extensive manual review methods,and lack scientific early warning methods of default risks,resulting in high default rates for borrowers.The excessive default rate has seriously affected the normal operation of the online loan platform,and at the same time increased the systemic financial risks in China.In this context,how to choose a scientific method for early warning of the default risk of the online loan platform and reduce the overall financial risk of the online loan industry has become a common concern of the academic community and the industry.The research purpose of this article is to compare the effectiveness of traditional statistical methods and machine learning algorithms in early warning of default risks on online loan platforms,to determine the default risk early warning tools that are most suitable for online loan platforms,and to provide a scientific decision basis for online loan platform loan review.At the same time,it solves the shortcomings of excessive default risk and high labor costs of online loan platforms,and provides support for the healthy and healthy development of online loan platforms.It can also indirectly reduce China's overall financial risk.On this basis,this article also points out a series of indicators that can be used as early warnings of default risks of online loan platforms,providing a reference basis for reducing financial risks of online loan platforms in practice.This article first introduces the development history of China's private lending and the process of turning into an online loan platform in the new century;then reviews the methods of domestic and foreign scholars' control of default risks on online loan platforms;and then introduces the risk control of domestic and foreign online loan platforms Method;since then,this article aims at big data risk control methods,and specifically introduces the theoretical basis and mathematical principles of various statistical models and machine learning algorithms;on this basis,this article uses photographs taken from May 2015 to February 2017 Lending real transaction data sets for model training and testing.Finally,the paper obtained the results of model training and found that the average accuracy rate of the machine learning integrated learning algorithm in the out-of-sample test set is as high as89.49%,which is significantly higher than the 60.94% accuracy rate of the Logistic model outside the sample.The ability of early warning of default risk is generally stronger than the Logistic model.And this article also compares inside the machine learning algorithm.It is found that the AdaBoost algorithm has the highest prediction accuracy,its accuracy outside the sample is 91.70%,and the training results in the training set are relatively stable.Therefore,this article recommends choosing machine learning The AdaBoost algorithm(the base classifier is a decision tree)deals with the early warning of default risks of online loan platforms.Finally,based on the significant results of the Logistic regression model,this paper obtains a series of early-warning indicators that are helpful for early warning of default risks of online loan platforms.It is recommended to combine the significant indicators obtained by the Logistic model with machine learning algorithms to collect users in practice.Corresponding index information,and early warning of default risk through trained machine learning algorithms.In addition,this paper also discusses the limitations of the insufficient accuracy of machine learning algorithms in distinguishing default samples.Although the ability of machine learning algorithms to distinguish default samples is better than the Logistic model,it can only identify about 13% of default samples.This article considers this to be the result of changes in the data characteristics of the default samples over time,so it is recommended that future research It further considers the impact of industry factors and macroeconomic background on the risk of default,and recommends that online loan platforms,online loan platforms and other institutions can share information with each other and provide users with more multi-dimensional data features to further improve machine learning.The out-of-sample prediction classification capability of the algorithm improves the early warning capability of the default risk of machine learning algorithms.In addition,this article also discusses the impact of the relatively simple time period of the data sample on the universality of the conclusions of this article.This article does not take into account the specific conditions of different stages of the development of the online loan platform.The statistical model and machine learning algorithm's default risk early warning effect may vary greatly in different time periods,which may affect the accuracy of the conclusions of this article.It will be further improved in future research.