摘要
随着大数据和机器学习的流行,其在破产预测和风险预测领域逐渐崭露头角。本文运用爬虫技术得到885家网贷平台的16815条数据,通过因子分析及模型验证挖掘出了若干能较好评估P2P平台风险的因子。然后本文通过选取的指标体系建立了Logistics回归、支持向量机、BP神经网络、LightGBM等单模型以及融合模型进行学习训练,所建立的融合模型在测试集中得到最高的准确率,说明本文所建的融合模型能较好地评估网贷平台的风险。本文还选取决策树绘图以及对特征进行重要性排名,选取出了对识别问题平台有重要作用的十项特征。这对投资者选取安全平台进行投资或者监管者选取重点平台进行监管有很好的借鉴意义。
Big data and machine learning have gradually emerged in the field of bankruptcyprediction and risk prediction.In this paper,python is used to obtain 16815 pieces of data from 885 online lending platforms.Through factor analysis and model verification,several characteristics that can better evaluate the risks of P2P platforms are mined.In addition,Logistics regression,support vector machine,BP neural network,LightGBM and fusion model are established through the selected indicator system for learning and training,and the fusion model obtains the highest accuracy in the test set,which shows that the fusion model established in this paper can better evaluate the risk of online lending platform.Then,we select one decision tree to draw and rank features in order of importance.We select ten features that are important for identifying problem platforms.Thispaperis of great reference significance fo rinvestors to choose safe platforms for investment or regulators to choose key platforms for supervision.
作者
梁爽
刁节文
肖邦
LIANG Shuang;DIAO Jie-wen;XIAO Bang(Schoolof Management,Shanghai Universityof Scienceand Technology,Shanghai 200093,China;Hanhai Information Technology(Shanghai)Co.,Ltd,Shanghai 200050,China)
出处
《运筹与管理》
CSSCI
CSCD
北大核心
2021年第1期170-176,共7页
Operations Research and Management Science
基金
上海市高原学科管理科学与工程资助(1018303010)。
