Research On Two-stage Feature Selection Method And Its Application In Enterprise Credit Risk Assessment

The financial crisis happened in 2008 has inflicted heavy losses on the global economy,under the trend of globalization,competition between companies is becoming more and more intense.As a result,enterprise credit risk has aroused widespread concern and the establishment of a sound risk prevention mechanism is imminent.The company has a variety of financial data.In the past,many credit risk models used this data to determine the credit risk of the company.However,in the context of big data,high-dimensional data has brought a lot of problems to modeling,the traditional credit risk assessment model is ineffective.As the same time,artificial intelligence has rapidly emerged.Support vector machine and other new technologies have been widely used in various fields,feature selection has become a powerful weapon for dimensionality reduction,and ensemble learning has reduced the bias and errors of individual classifiers by merging multiple sub-classifiers.Based on this background,this paper proposes a two-stage feature selection method in academic research.Firstly,the feature selection methods are tested for stability.The stability test ensures that the selected features are representative of the entire data set.Then,in the past research,the number of optimal features is often determined by experience.This paper quantifies each feature and introduces the idea of wrapper feature selection into filter feature selection to achieve removal of redundancy,reduce dimensions and improve model accuracy.Finally,based on the work of two-stage feature selection,the hybrid model HFMG is proposed,and a combination of multiple sub-learners is adopted to further improve the classification ability and enhance the model reliability.This article combines the credit risk evaluation of listed companies to do empirical analysis.The real data set includes 160 listed companies.Each company has 22 characteristic attributes.Experimental results show that the studying of the two-stage feature selection method is of great significance.The hybrid model HFMG has achieved good results,thus,it can not only be used in credit risk assessment,but also can be applied to more other fields.