基于多分类器动态集成的P2P违约风险评估  

P2P Default Risk Assessment Based on the Dynamic Aggregation of Multiple Classifiers

作　　者：胡忠义 王超群 陈远[1,2] 吴江[1] 鲍玉昆[3] HU Zhongyi;WANG Chaoqun;CHEN Yuan;WU Jiang;BAO Yukun(Wuhan University,Wuhan,China;Huazhong University of Science and Technology,Wuhan,China)

机构地区：[1]武汉大学信息管理学院 [2]武汉大学信息资源研究中心 [3]华中科技大学管理学院

出　　处：《管理学报》2019年第6期915-922,共8页Chinese Journal of Management

基　　金：国家自然科学基金资助重大项目(91646206);国家自然科学基金资助项目(71573197);国家自然科学基金资助青年项目(71601147)

摘　　要：针对P2P违约风险评估问题,构建基于K均值聚类的多分类器动态集成模型。通过K均值聚类将P2P借贷样本集划分为多个区域,并在每个区域分别构建多个基分类器模型,进行多分类器集成;在给定新的测试样本时,基于样本特征动态选择样本归属区域的基分类器,进行模型集成与违约风险评估。以Lending Club借款数据为例,对网贷违约风险评估进行实证分析表明,基于多分类器动态集成的P2P违约评估模型,比单一模型和静态集成模型的准确性更好,且基于神经网络的动态集成模型效果最好。To evaluate the personal default risk on P2P lending platforms,this study proposes a multiple classifier dynamic aggregation model based on K-means clustering.The samples of P2P lending are divided into several subgroups by means of K-means clustering,and multiple classifiers are built for model aggregation in each subgroup.Given new testing samples,the corresponding multiple classifiers and aggregated model are dynamically selected based on the similarity between the testing samples and subgroups to assess the default risk.Taking dataset from Lending Club as a case of P2P personal credit scoring,the experimental results show that the proposed dynamic aggregation model outperforms both single and static aggregation models;additionally,dynamic aggregation model based on neural network is the best amongst examined models.

关 键 词：P2P网络借贷 多分类器集成 动态选择 K均值聚类 

分 类 号：C93[经济管理—管理学]