基于Multiple-Lasso-Logistic回归模型的车险索赔概率预测

Prediction of Automobile Insurance Claim Probability Based on Multiple-Lasso-Logistic Regression Model

摘要: 近年来，Logistic回归模型在非寿险精算科学中得到了广泛的应用，本文对法国的一组车险索赔数据，采用Lasso及其扩展方法结合Logistic模型建立车险索赔预测模型，同时引入了惩罚权重，并与Lasso-logistic回归模型和Logistic回归模型进行比较，结果表明：模型综合性能最优的是multiple-Lasso- Logistic回归模型。并在此基础上筛选出了预测性能最强的因子水平，同时对响应变量作用相同的水平进行了融合，有效地降低了变量维度。

Abstract: In recent years, Logistic regression model has been widely used in non-life insurance actuarial sci-ence. This paper uses Lasso and its extension method combined with Logistic regression model to establish a prediction model for vehicle insurance claims based on a group of vehicle insurance claim data in France, and introduces penalty weight. Compared with Lasso-logistic regression model and Logistic regression model, the results show that the multiple-Lasso-Logistic regression model has the best comprehensive performance. On this basis, the factor level with the strongest predic-tion performance is selected, and the level with the same effect of response variables is fused, which effectively reduces the variable dimension.