摘要
近年来,个人信用评估问题成为信贷行业的研究热点,针对当前应用于信用评估的分类算法大多存在只对某种类型的信用数据集具有较好的分类效果的问题,提出了基于Gradient Boosted Decision Tree(GBDT)的个人信用评估方法。GBDT天然可处理混合数据类型的数据集,可以发现多种有区分性的特征以及特征组合,不需要做复杂的特征变换,对于特征类型复杂的信用数据集有明显的优势,且其通过其损失函数可以很好地处理异常点。在基于两个UCI公开信用审核数据集上的对比实验表明,GBDT明显优于传统常用的支持向量机(Support Vector Machine,SVM)以及逻辑回归(Logistic Regression,LR)的信用评估效果,具有较好的稳定性和普适性。
In recent years, the personal credit scoring problem has become the research hotspots in the credit industry. In view of the current classification algorithms applied in credit scoring only have a good effect for some type of credit data set, a personal credit scoring method based on gradient boosted decision tree （GBDT） methods is put forward in this paper. GBDT is naturally able to deal with mixed types of data sets and find distinguishing features and feature combinations without doing complex feature transformation. GBDT shows obvious advantages for credit data set of complex data types, and by the loss function outliers can be well processed. The contrast experiment based on two UCI public credit audit data sets shows that credit scoring results of GBDT is obviously superior to the result of Support Vector Machine （SVM） and Logistic Regression （LR） with good stability and universal applicability.
出处
《电子设计工程》
2017年第15期68-72,共5页
Electronic Design Engineering
关键词
信用评估
分类算法
GBDT
GBDT
credit scoring
classification algorithms
GBDT
