基于超参数优化和集成算法的个人信用评估研究  

Research on Personal Credit Evaluation Based on Super Parameter Optimization and Integration Algorithm

作　　者：林卫明 曹杰 LIN Weiming;CAO Jie(School of mathematics and statistics,Nanjing University of Information Science & Technology,Nanjing 210044,China;Xuzhou Institute of Technology, Xuzhou 221018,China)

机构地区：[1]南京信息工程大学数学与统计学院,江苏南京210044 [2]徐州工程学院,江苏徐州221018

出　　处：《徐州工程学院学报（自然科学版）》2020年第4期67-74,共8页Journal of Xuzhou Institute of Technology(Natural Sciences Edition)

摘　　要：为提高信用特征集与目标标签的相关性,降低特征之间的冗余性,采用Person相关系数、卡方、递归特征消除、Lasso回归和随机森林5种特征选择方法投票表决,构造新的特征子集;为进一步提升模型预测效果,选用Hyperband算法对RF、XGboost集成模型参数优化,优化之后的hp-RF、hp-XGboost在Loan数据集上进行五折交叉验证,实证具有更好的识别效果,并在P2P人人贷数据上进行了应用,为我国信用评估的构建提出了新的方式.In order to improve the correlation between the credit feature set and the target tag and reduce the redundancy between the features,five feature selection methods,namely person correlation coefficient,chi square,recursive feature elimination,Lasso regression and random forest,are used to construct a new feature subset.To further improve the prediction effect of the model,the hyper band algorithm is used to optimize the parameters of RF and xgboost integrated models,After optimization,hp-RF and hp-XGboost are cross validated with 50%discount on loan data set,and the empirical results have better recognition effect.They are applied to P2P Renren loan data,which provides a new way for the construction of credit evaluation in China.

关 键 词：特征选择 Hyperband算法 交叉验证 

分 类 号：F832.4[经济管理—金融学]