特征选择方法在信用评估指标选取中的应用

在信用评分模型中所运用的指标变量对模型的表现有重要的影响,指标选取方法的科学化规范化水平有待于进一步提高。本文研究了机器学习领域的特征选择方法在定量确定信用评分模型指标体系上的应用。以实际信用评估问题为例,对四种特征选择方法(ReliefF方法、基于相关性的方法、基于一致性的方法和包裹性)进行了比较试验,验证了特征选择方法可以在精简性、速度和准确率三个方面提高信用评分模型的表现。其中基于一致性的方法和包裹法表现优于Reli-efF方法和基于相关性的方法。

The Application of Feature Selection Methods in Variables Selections For Credit Scoring Models

The variables included in credit scoring models have important impacts on models' performance,but the variable selevtion methods nee4d to be more standardized and systemized.This paper presents an empirical study on the application of feature selection methods in variables selections compare four feature selection methods-"ReliefF","Correlation-based","Consistency-based"and "Wrapper"methods.The study illustrates how these methods help to improve the performance of scoring models in three aspects:model simplicity,model speed and model accuracy.The results show that "Consistency-based"and "Wrapper"methods perform better than "ReliefF"and "Correlation-based" methods.