现代信息科技2025，Vol.9Issue(19)：31-37,42,8.DOI:10.19850/j.cnki.2096-4706.2025.19.007
基于机器学习的车险欺诈检测方法研究
Research on Auto Insurance Fraud Detection Method Based on Machine Learning
文晓帅 1闫方正 1邹印 1程一帆 1李雨擎 1戴妍玲1
作者信息
- 1. 中国矿业大学(北京)人工智能学院,北京 100083
- 折叠
摘要
Abstract
Automatic detection of potential auto insurance fraud can maintain the market order of the auto insurance industry.At present,the traditional Machine Learning methods face the challenges of small data volume,unbalanced category distribution and feature redundancy in implementing auto insurance fraud detection.Aiming at these problems,this paper selects LightGBM,XGBoost,AdaBoost and Random Forest models,and adopts a set of methods including data preprocessing,feature selection,hyperparameter optimization and category weight adjustment.The experimental results show that the highest ROC-AUC values of each model are 0.906 9,0.840 7,0.846 7 and 0.851 3,respectively.At the same time,through the collaborative analysis based on feature frequency and decision contribution,the importance of key features is discussed,and the method is applied to the practical scenarios.The research shows that this method can effectively improve the accuracy and model interpretability of auto insurance fraud detection,and provides a more reliable anti-fraud tool for the insurance industry.关键词
机器学习/特征选择/LightGBM/车险欺诈检测Key words
Machine Learning/feature selection/LightGBM/auto insurance fraud detection分类
信息技术与安全科学引用本文复制引用
文晓帅,闫方正,邹印,程一帆,李雨擎,戴妍玲..基于机器学习的车险欺诈检测方法研究[J].现代信息科技,2025,9(19):31-37,42,8.
