Research On Credit Risk Assessment Based On Interpretable Fusion Algorithm Model

With the prosperity and development of China's market economy,our national credit system development strategy has been officially put forward.The growing demand for a better life of the people also requires banks,loan platforms and other financial institutions to constantly adapt to the market requirements and provide users with more extensive and accurate business support.At the same time,malicious lending and other acts emerge one after another,causing heavy losses to financial institutions.Therefore,banks,loan platforms and other financial institutions turn their attention to the credit evaluation system,and correctly judge the borrower's ability,willingness and final repayment result has become a hot topic.This thesis mainly studies the loan problem in the credit evaluation system,and studies and improves the impact of credit information on loans.At present,the credit evaluation system widely used in banks,loan platforms and other financial institutions is a traditional expert model constructed by statistical methods.In the era of big data,such models are gradually inadequate,and there are some problems,such as long time-consuming,poor accuracy and unable to process a variety of data.This paper mainly makes the following improvements:Firstly,the fusion algorithm model is improved based on stacking algorithm.Through the public data set of lending club provided on kaggle website for data analysis,data visualization,outlier missing value processing,regularization and normalization,the fusion model based on stacking algorithm is established and compared with a variety of common machine learning algorithm models,The experimental analysis shows that the improved stacking algorithm fusion algorithm model has the best classification performance and effect and the highest efficiency under the comprehensive evaluation of various classification indexes.Finally,aiming at the problem of weak interpretability of machine learning model and other complex learning models,the local interpretability algorithm based on differential optimization is applied to the credit evaluation model.Through the interpretation of people's common understanding methods,the attempt to explain the problem of machine learning black box model is realized,and the reliability and security of the model are enhanced,Experiments show that the improved local interpretability algorithm has good fitting effect and stability in time span.