Research On XGBoost Credit Assessment Model For Small And Micro Enterprises Based On Gibbs Sampling

In the context of big data,applying machine learning to the credit assessment field has become a research focus,and exploring more effective feature screening methods through the combination of big data and machine learning is of practical significance for building credit assessment models for small and micro enterprises.The main focus of this paper is to use Gibbs sampling to screen out the characteristic variables that are more strongly associated with credit assessment for XGBoost model construction and to analyze the model explanatory power of the characteristic variables using the SHAP explanatory model.Feature screening is the key to constructing machine learning models.There are two main problems in feature screening for high-dimensional feature spaces,one is that traditional feature screening methods cannot handle nonlinear correlations between feature variables;the other is that traversing the feature space has exponential computational complexity and is computationally infeasible.Therefore,the use of Gibbs sampling-based feature screening method extracts more effective feature variables while reducing the computational complexity.The results of the empirical analysis in this paper show that the XGBoost model constructed based on Gibbs sampling performs better in terms of model results.Since machine learning models are black box models,there is a need to make the complex models transparent.In this paper,we use SHAP explanation model to explain and analyze the XGBoost model based on Gibbs sampling,and investigate the screened feature variables both globally and locally to verify the effectiveness of the Gibbs sampling based feature screening method.