融合贷款信息和轨迹特征的催收评分卡模型
Collection Scorecard Model Integrating Loan Information and Trajectory Features
王延松 1张华伟 1张磊 1王蒙 1陈剑2
作者信息
- 1. 奇瑞徽银汽车金融股份有限公司 芜湖 241000
- 2. 长三角信息智能创新研究院 芜湖 241000
- 折叠
摘要
针对现有汽车金融中车辆贷后逾期风险识别准确率低、效率低等问题,论文提出了一种XGBoost-SMOTE的逾期风险预测模型.该模型通过采用SMOTE的过采样方法构建相对均衡的不同类别比例数据集,提高了模型的泛化能力.同时,结合了车辆用户基本信息、贷款信息和车辆GPS轨迹信息来进行特征工程,将融合后的特征数据输入到XGBoost分类器中,最终完成了车辆贷后逾期风险识别,实验结果表明:相较于其他机器学习算法预测模型,该模型可以达到更好的预测性能.
Abstract
This paper proposes a post-loan overdue risk of vehicle prediction model based on XGBoost-SMOTE to address the problems of the existing monitoring methods which can not identify risks effectively.In this paper,over sampling method of SMOTE is used to build a relatively balanced positive and negative proportion data set,which improves the generalization ability of the mod-el.And combined with the basic information of vehicle users,loan information and vehicle GPS track information for feature engi-neering.Then,the integrated feature data is input into XGBoost classifier,and finally complete the identification of post-loan over-due risk of vehicle.Compared with other machine learning algorithm prediction models,this model can achieve better prediction per-formance after the experiment.
关键词
汽车金融/贷后逾期/GPS轨迹/风险特征知识Key words
automobile finance/post-loan overdue/GPS trajectory/risk characteristics knowledge引用本文复制引用
基金项目
2021年安徽省重点研究与开发计划(202104a05020071)
出版年
2024
国家科技期刊平台
NSTL
万方数据