计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (5): 34-49.DOI: 10.3778/j.issn.1002-8331.2109-0312

• 热点与综述 • 上一篇    下一篇

机器学习模型在车险欺诈检测的研究进展

卢冰洁,李炜卓,那崇宁,牛作尧,陈奎   

  1. 1.之江实验室,杭州 311121
    2.南京邮电大学 现代邮政学院,南京 210003 
    3.东南大学 苏州联合研究生院,江苏 苏州 215123 
    4.南京大学 计算机软件新技术国家重点实验室,南京 210093
  • 出版日期:2022-03-01 发布日期:2022-03-01

Survey of Auto Insurance Fraud Detection with Machine Learning Models

LU Bingjie, LI Weizhuo, NA Chongning, NIU Zuoyao, CHEN Kui   

  1. 1.Zhejiang Lab, Hangzhou 311121, China
    2.School of Modern Posts, Nanjing University of Posts and Telecommunications, Nanjing 210003, China
    3.Southeast University-Monash University Joint Graduate School, Southeast University, Suzhou, Jiangsu 215123, China
    4.State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210093, China
  • Online:2022-03-01 Published:2022-03-01

摘要: 随着保险行业的蓬勃发展,保险欺诈问题也显得日趋严重。车险欺诈一直是保险欺诈的“重灾区”,对保险行业的发展至关重要。因此,车险欺诈检测技术一直是国内外学者研究的热点问题。鉴于我国在机动车辆保险欺诈检测技术方相对滞后,而国外的研究成果又较少对我国车险业务数据进行有效建模与分析,首次针对机器学习模型应用在车险欺诈检测的研究工作进行了文献调研,对二十多年来的研究工作进行系统化的归纳与总结。通过引入车险欺诈流程的简介,对专家系统与智能理赔系统在车险欺诈检测的流程进行了叙述;依次从国外和国内的角度介绍了机器学习模型应用在车险欺诈检测的具体研究进展,并进行了宏观的对比;基于国内某车险公司提供近5年来高质量的车险数据选取具有代表性的机器学习模型进行建模,并进行了全面的测试与分析;探讨了车险欺诈检测技术未来的研究方向。

关键词: 汽车保险欺诈, 机器学习, 深度学习, 数据不均衡, 保险监管

Abstract: With the vigorous development of insurance industry, the problem of insurance fraud has become increasingly serious. Auto insurance fraud has been the hardest “hit area” of insurance fraud, which is very important for the development of insurance industry. Therefore, auto insurance fraud detection technology has been a hot top for researchers. Considered that the fraud detection techniques of automobiles in China have been lagged, and research results in aboard are not enough to model and analyze China’s auto insurance business data, this paper presents the first work on the survey of the methods of machine learning applied in auto insurance fraud detection, and systematically summarizes the related works in the two decades. It firstly gives a brief introduction to the process of auto insurance fraud detection based on expert systems and intelligent claim settlement systems. Then, it describes the specific research progresses of machine learning models applied in auto insurance fraud detection at home and aboard. Moreover, with the high-quality auto insurance data provided by the insurance company in recent five years, it carries out a detailed comparison based on the representative models from machine learning and gives a comprehensive analysis. Finally, the future research direction of auto insurance fraud detection is discussed and outlooked.

Key words: auto insurance fraud, machine learning, deep learning, unbalanced data, insurance regulation