%0 Journal Article
%T 基于Stacking模型融合的车辆保险反欺诈识别<br>Anti-Fraud Identification of Vehicle Insurance Based on Stacking Model Fusion
%A 金天颖
%A 张美辰
%A 王可欣
%J Statistics and Applications
%P 486-495
%@ 2325-226X
%D 2024
%I Hans Publishing
%R 10.12677/sa.2024.132048
%X 基于Kaggle公开数据集，首先对其进行了数据清洗、数据均衡化等预处理工作，确保数据的质量和适用性。在预处理完成后，使用随机森林、XGBoost和LightGBM三种不同的机器学习算法构建了保险欺诈识别模型，并通过网格搜索法调整参数来提高其性能。采用分类模型的评估方法，对这三个模型的Pression、Recall等指标进行了对比分析，结果显示LightGBM模型在整体的分类效果上表现最好。最后，我们引入了Stacking技术，将三个单一模型作为初级分类器，Logistic Regression作为元分类器，得到了融合的的车险欺诈识别模型。这一模型结合了三个单一模型的优点，不仅具有较高的稳定性，还能提高整体的预测精度。通过模型融合，我们得到了一个更加全面、准确的欺诈识别系统，为保险公司提供了更可靠的风险管理工具。<br />
Based on the Kaggle public dataset, we first conducted preprocessing such as data cleaning and balancing to ensure the quality and applicability of the data. After preprocessing, we constructed insurance fraud detection models using three different machine learning algorithms: Random Forest, XGBoost, and LightGBM. We used grid search to tune the parameters to improve their performance. By evaluating the models using classification metrics such as Precision and Recall, we found that the LightGBM model performed the best overall. Finally, we introduced the Stacking technique, combining the three individual models as base classifiers and Logistic Regression as the meta-classifier, to obtain a fused auto insurance fraud detection model. This model combines the advantages of the three individual models, exhibiting not only high stability but also improved overall prediction accuracy. Through model fusion, we obtained a more comprehensive and accurate fraud detection system, providing insurance companies with more reliable risk management tools.
%K 模型融合，XGboost，LightGBM，随机森林，车险欺诈<br>Model Fusion
%K XGBoost
%K Lightgbm
%K Random Forest
%K Automobile Insurance Fraud
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=85411