摘要
针对网络现金贷中存在的金融机构数据量匮乏、数据不共享和第三方机构单独建模等问题,本文提出一种基于SMOTETomk采样和迁移学习的违约分类算法。首先,由于数据存在不平衡特点,通过SMOTETomk算法进行综合采样;其次,用特征提取和微调两种方法构建迁移学习模型;最后将本文模型和logistic模型进行性能参数对比,结果表明:本文分类器有更高的评价结果。
Aiming at the lack of data in online cash loans, the lack of data sharing, and the independent modeling of thirdparty institutions, this paper proposes a default classification algorithm based on SMOTETomk sampling and transfer learning.Firstly, SMOTETomk is designed to acquire sample data in order to solve the problem of imbalance. Secondly, two neural network models based on transfer learning are constructed by feature extraction and fine tuning. Finally, the model in this paper is compared with the logistic model. The results displayed that the default prediction model established by transfer learning has better evaluation result.
作者
杨冰清
赵金虎
YANG Bingqing;ZHAO Jinhu(School of Mathematics and Statistic,Fuyang Normal University,Fuyang Anhui 236037,China)
出处
《阜阳师范大学学报(自然科学版)》
2022年第3期6-11,共6页
Journal of Fuyang Normal University:Natural Science
基金
安徽省高校自然科学研究项目(KJ2020A0540)资助。
关键词
迁移学习
违约预测模型
不平衡数据
现金贷
transfer learning
default prediction model
imbalanced data
cash loans