Research On Machine Learning Methods For Interpretable Credit Risk Assessment

Credit risk assessment is a crucial component of enterprise risk management.The rise of artificial intelligence technology and machine learning methods offer new ways to enhance the intelligence and efficiency of credit risk assessment.However,methods such as neural networks and deep learning present challenges,including poor interpretability and low data transparency,which discourage their adoption in enterprise credit risk assessment.To address these issues and enhance the interpretability and intelligence of credit risk assessment,this paper introduces a new method based on ensemble learning.This innovative approach offers robust interpretability tailored to credit risk assessment and furthers research and practical application in this area.The key research components and novel contributions of this paper are as follows:1.Addressing the issue of poor interpretability in credit risk assessment,and building upon domestic and international research,this paper introduces a new credit risk assessment method using an ensemble learning approach.Initially,inspired by the concept of XGBoost(e Xtreme Gradient Boosting),the new method incorporates a tree splitting method without repeated paths,generating a set of classification trees from the original dataset.Subsequently,linear logistic regression learning is performed on these datasets.Ultimately,the Ada Boost(Adaptive Boosting)method,with logistic regression as the base model,is employed to linearly weight and ensemble the logistic regression models on different classification trees,forming a fully interpretable credit risk assessment model.The new model’s distinctive features allow it to serve as an effective,interpretable reference for enterprise credit risk management.In terms of model performance,the accuracy and recall rates on the Beijing enterprise audit dataset are 93.10% and 90.51% respectively,and on a desensitized real dataset of a city,they reach 82.68% and 83.47% respectively.These results are comparable to the performance of existing credit risk assessment models.Through mathematical derivation and experimental verification,it is demonstrated that the new model possesses excellent interpretability and readability,offering a valuable reference for credit risk assessment.Consequently,this model holds significant practical value.2.Building upon the new logistic regression integration model,we have designed and implemented an enterprise credit risk assessment system.This system enables enterprise credit data processing and analysis,and constructs a comprehensive and efficient credit risk assessment workflow.The system’s primary modules include user management,data management,data preprocessing,and risk assessment.The user management module provides user permissions and information query functions.The data management module offers enterprise credit data management and basic dataset display functionalities.The data preprocessing module implements various data preprocessing methods to complete data cleaning and filling.The risk assessment module integrates the newly proposed method with other mainstream machine learning methods to evaluate credit risk,and offers a comparative analysis of the model’s interpretability.Tested with actual data,the results demonstrate that the credit risk assessment system is functional and performs well.It also provides intuitive chart displays and comparative analysis between different models.In conclusion,the interpretable credit risk assessment model proposed in this paper offers a novel solution with enhanced interpretability for enterprise credit risk assessment.It serves as a valuable reference for applying machine learning methods to credit risk assessment problems.The successful implementation of the aforementioned system and the test results demonstrate that the newly proposed model can be effectively applied to actual enterprise credit risk assessment scenarios,affirming its clear practical value.