Research on Computer Intelligent Risk Prediction Model and Identification Algorithm with Machine Learning

Abstract

With the rapid development of economics, consumer lending and trust loan have gained unprecedented growth all over the world. Due to the serious consequence of borrowers' breach of contract, banks need a valid method to help managing credit risk. Till now, the most common way is to set up blacklist and whitelist according to previous loan records to assist bank in deciding whether to offer loan to someone or not. In other words, there is a fixed threshold to filter customers. Once borrower has a number of violations beyond the threshold, the bank will add him into the blacklist which means the man cannot get debt in the future according to his poor credit. This simple method may be useful but it also has some inevitable downsides resulting from its empirical nature. For instance, this approach still needs to be conducted with countless bank workers, accounting for its failure in satisfying soaring need in current changeable loan scenario. Meanwhile, the manually set threshold is not sensitive to massive data, which is more than unavoidable in today's life. To the best of our knowledge, machine learning methods are suitable to deal with huge amount of data and make accurate predictions. Considering the characteristics in consumer lending and trust loan, we put forward a new algorithm model implemented with machine learning knowledge to determine whether or not a loan should be granted. The method contains some cutting-edge machine learning ideas such as tree model and neural network. This new credit scoring framework can solve above problems effectively and control the expenditure within an acceptable range. In the experiments, we evaluate our method extensively on bank credit dataset, and the results demonstrate that it outperforms most credit scoring and risk prediction methods, achieving an AUC of 0.840 on the Test Set.