随着中国新能源汽车的兴起,关于汽车保险诈骗的问题日益突出。为了对保险诈骗行为进行有效识别,本文基于机器学习的相关理论,利用模拟退火算法调参的Stacking融合模型对保险欺诈进行预测。首先,利用随机森林和XGBoost算法筛选得到两个不同特征的训练数据集,然后通过差异化的数据来优化Stacking模型的预测能力,并利用交叉验证法得到最优模型,其准确率为87.43%。实证分析表明,相较于未使用差异化数据的Stacking模型,本文所建的融合模型对汽车保险欺诈行为有更高的识别能力。With the rise of new energy vehicles in China, the issue of car insurance fraud has become increasingly prominent. In order to effectively identify fraudulent insurance activities, this study employs the Stacking ensemble model, optimized using simulated annealing algorithm tuning based on machine learning theories, to predict insurance fraud. Initially, utilizing the Random Forest and XGBoost algorithms, two distinct feature sets are selected to construct training datasets. Subsequently, by employing differentiated data, the predictive capability of the Stacking model is enhanced. Through cross-validation, the optimal model is obtained and its accuracy is 87.43%. Empirical analysis shows that compared to the Stacking model without differentiated data, the ensemble model developed in this study exhibits superior capability in identifying fraudulent behaviors in car insurance.
