动态异质集成信用评分模型在P2P网络借贷中的应用  

An Application of Dynamic Heterogeneous Ensemble Credit Scoring Model to Peer-to-Peer Lending

作　　者：刘传哲[1] 马达亮 夏雨霏 Liu Chuanzhe;Ma Daliang;Xia Yufei(School of Management,China University of Mining and Technology,Jiangsu Xuzhou 221116)

机构地区：[1]中国矿业大学管理学院,江苏徐州221116

出　　处：《金融发展研究》2018年第9期24-31,共8页Journal Of Financial Development Research

基　　金：中央高校基本科研业务费专项资金资助(2017BSCXB20)

摘　　要：本文借鉴了传统信用评分方法,提出了适用于P2P网络借贷环境的动态异质集成分类模型DSHE。该模型能够实现对冗余特征变量的筛选,具有一定的高维数据处理能力;其异质集成结构与动态筛选策略能够实现基础模型权重的自适应调整,从而提高信用评估性能。使用UCI数据库中的数据和网贷真实数据进行实证分析,结果表明,异质集成模型整体表现较优;DSHE模型在预测准确率上表现突出,在4个评价指标下的平均秩优于Logistic回归等对比模型。Following the mechanism of traditional credit scoring methods,a novel dynamic selective heterogeneous ensemble(DSEH)model suitable to the application of P2P lending is proposed.The model provides a feature selection algorithm,which is able to filter redundant features and handle high-dimension data.The heterogeneous structure and dynamic selection strategy can adaptively adjust the weights of base models and thus,enhance the evaluation capability.UCI credit dataset and real dataset from two P2P lending platforms are used to validate the proposal.The results show that DSHE outperforms other mainstream credit scoring approaches in several measures.The average rank of proposed DSHE is superior to baseline models including logistic regression.

关 键 词：信用评分 动态异质集成 P2P网络借贷 

分 类 号：F832[经济管理—金融学]