保险大数据条件下车险费率厘定的研究——基于SOM神经网络方法的车险索赔强度建模
|
摘要
神经网络是近年来机器学习领域的研究热点之一。该方法在许多领域都有成功的应用,但较少应用于汽车保险索赔预测中。本研究将自组织竞争神经网络(SOM)应用于汽车保险的索赔预测中,在此基础上建立车险索赔强度模型。本研究将影响车险索赔的因素分为三类:从人因素、从车因素、地域因素。对于从车因素,通过应用SOM神经网络方法对多个解释变量进行聚类分析来获得综合影响评价指标——从车因子综合变量。进一步按照索赔强度的高低,将该变量分成5个水平,进而起到减少解释变量的作用。将地域因素作为随机效应,以从人因素变量和从车因子综合变量为自变量,以索赔强度为因变量,建立广义线性混合模型。本文创新在于:在充分考虑了影响车险费率的各种因素下,应用SOM神经网络聚类方法减少自变量的个数,为车险费率厘定提供了一种新思路。
Neural network is one of the research hotspots in the field of machine learning in recent years. This method has been successfully applied in many fields,however it is seldom used in automobile insurance claim prediction. In this paper,the self-organizing competitive neural network( SOM) was applied to the prediction of automobile insurance losses,and the claim severity model of automobile insurance was established. The paper divided the factors affecting automobile insurance claims into three categories: human-related factors,automobile-related factors and region-related factors. The multiple explanatory variables were clustered by SOM method to obtain the comprehensive variables of the automobile-related factors. According to the extent of claim severity,these variables were then divided into five levels,which could reduce the explanatory variables. Region-related factors were used as random effects,and the comprehensive variables of automobile-related factors and human-related factors were introduced into the model as independent variables. The claim severity was regarded as dependent variable.The generalized linear mixed model was thus established. One of the advantages of this paper is that,while taking into account all factors affecting automobile insurance rate making,the SOM clustering method reduces the number of independent variables,which provides a new vein of thought for automobile insurance claim pricing.
Neural network is one of the research hotspots in the field of machine learning in recent years. This method has been successfully applied in many fields,however it is seldom used in automobile insurance claim prediction. In this paper,the self-organizing competitive neural network( SOM) was applied to the prediction of automobile insurance losses,and the claim severity model of automobile insurance was established. The paper divided the factors affecting automobile insurance claims into three categories: human-related factors,automobile-related factors and region-related factors. The multiple explanatory variables were clustered by SOM method to obtain the comprehensive variables of the automobile-related factors. According to the extent of claim severity,these variables were then divided into five levels,which could reduce the explanatory variables. Region-related factors were used as random effects,and the comprehensive variables of automobile-related factors and human-related factors were introduced into the model as independent variables. The claim severity was regarded as dependent variable.The generalized linear mixed model was thus established. One of the advantages of this paper is that,while taking into account all factors affecting automobile insurance rate making,the SOM clustering method reduces the number of independent variables,which provides a new vein of thought for automobile insurance claim pricing.
引文
[1]傅鸿源,姚尧,李良.基于RBF网络的工程保险费率厘定研究[J].系统工程理论与实践,2008,28(7):169-172.
[2]库恩,约翰逊.应用预测建模[M].林荟,等译.北京:机械工程出版社,2016:20-43.
[3]孟生旺.神经网络模型与车险索赔频率预测[J].统计研究,2012,29(03):22-26.
[4]孟生旺,徐昕.非寿险费率厘定的索赔频率预测模型及其应用[J].统计与信息论坛,2012,7(9):14-19.
[5]孙维伟,张连增. ZAIG模型在车险定价中的应用研究[J].保险研究,2013,(4):43-51.
[6]王新军,王亚娟.基于广义线性模型的车险分类费率厘定研究[J].保险研究,2013,(9):43-65,85.
[7]叶明华.基于BP神经网络的保险欺诈识别研究——以中国机动车保险索赔为例[J].保险研究,2011,(3):79-86.
[8]张连增,孙维伟,段白鸽. GLM与GAM在车险索赔频率建模中的应用及其比较[J].天津财经大学学报,2012,12:47-56.
[9]张连增,孙维伟.广义线性混合模型在保险索赔中的应用及R实现[J].江西大学财经学报,2013,(4):48-58.
[10]周志华.机器学习[M].北京:清华大学出版社,2016:97-119.
[11] Brockett P L,Xia X,Derrig R A. Using Kohonen’s Self-Organizing Feature Map to Uncover Automobile Bodily Injury Claims Fraud[J]. The Journal of Risk and Insurance,1998,65(2):245-274.
[12] Frees E W,Meyers G,Derrig R A(eds.). Predictive Modeling Applications in Actuarial Science:Volume 2,Case Studies in Insurance[M]. Cambridge University Press,New York,2016.
[13] Gao G,Wüthrich M V. Feature Extraction from Telematics Car Driving Heatmaps[J]. https://ssrn. com/abstract=3070069,2017.
[14] Hastie T,Tibshirani R,Friedman J. The Elements of Statistical Learning[M]. 2nd Ed. Springer,New York,2009.
[15] Lee S,Lin S,Antonio K. Delta Boosting Machine and Its Application in Actuarial Modeling. ASTIN,AFIR/ERM and IACA Colloquia,Institute of Actuaries of Australia,2015:1-21.
[16] Liu Y,Wang B,Lv S. Using Multi-class Adaboost Tree for Prediction Frequency of Auto Insurance[J].Journal of Applied Finance and Banking,2014,4(5):45-53.
[17] Wehrens R,Buydens L. Self-and Super-organizing Maps in R:The Kohonen Package[J]. Journal of Statistical Software,2007,21(5):1-18.
[18] Tornike M. Vehicle Insurance Claim Data Study and Forecasting Model Using Artificial Neural Networks[D]. Tallinn University of Technology,Estonia,2016.
[19] Wüthrich M V. Covariate Selection from Telematics Car Driving Data[J]. European Actuarial Journal 2017,7,89-108.
[20] Wüthrich M V,Buser C. Data Analytics for Non-life Insurance Pricing[J]. https://ssrn. com/abstract=2870308,2018.
[2]库恩,约翰逊.应用预测建模[M].林荟,等译.北京:机械工程出版社,2016:20-43.
[3]孟生旺.神经网络模型与车险索赔频率预测[J].统计研究,2012,29(03):22-26.
[4]孟生旺,徐昕.非寿险费率厘定的索赔频率预测模型及其应用[J].统计与信息论坛,2012,7(9):14-19.
[5]孙维伟,张连增. ZAIG模型在车险定价中的应用研究[J].保险研究,2013,(4):43-51.
[6]王新军,王亚娟.基于广义线性模型的车险分类费率厘定研究[J].保险研究,2013,(9):43-65,85.
[7]叶明华.基于BP神经网络的保险欺诈识别研究——以中国机动车保险索赔为例[J].保险研究,2011,(3):79-86.
[8]张连增,孙维伟,段白鸽. GLM与GAM在车险索赔频率建模中的应用及其比较[J].天津财经大学学报,2012,12:47-56.
[9]张连增,孙维伟.广义线性混合模型在保险索赔中的应用及R实现[J].江西大学财经学报,2013,(4):48-58.
[10]周志华.机器学习[M].北京:清华大学出版社,2016:97-119.
[11] Brockett P L,Xia X,Derrig R A. Using Kohonen’s Self-Organizing Feature Map to Uncover Automobile Bodily Injury Claims Fraud[J]. The Journal of Risk and Insurance,1998,65(2):245-274.
[12] Frees E W,Meyers G,Derrig R A(eds.). Predictive Modeling Applications in Actuarial Science:Volume 2,Case Studies in Insurance[M]. Cambridge University Press,New York,2016.
[13] Gao G,Wüthrich M V. Feature Extraction from Telematics Car Driving Heatmaps[J]. https://ssrn. com/abstract=3070069,2017.
[14] Hastie T,Tibshirani R,Friedman J. The Elements of Statistical Learning[M]. 2nd Ed. Springer,New York,2009.
[15] Lee S,Lin S,Antonio K. Delta Boosting Machine and Its Application in Actuarial Modeling. ASTIN,AFIR/ERM and IACA Colloquia,Institute of Actuaries of Australia,2015:1-21.
[16] Liu Y,Wang B,Lv S. Using Multi-class Adaboost Tree for Prediction Frequency of Auto Insurance[J].Journal of Applied Finance and Banking,2014,4(5):45-53.
[17] Wehrens R,Buydens L. Self-and Super-organizing Maps in R:The Kohonen Package[J]. Journal of Statistical Software,2007,21(5):1-18.
[18] Tornike M. Vehicle Insurance Claim Data Study and Forecasting Model Using Artificial Neural Networks[D]. Tallinn University of Technology,Estonia,2016.
[19] Wüthrich M V. Covariate Selection from Telematics Car Driving Data[J]. European Actuarial Journal 2017,7,89-108.
[20] Wüthrich M V,Buser C. Data Analytics for Non-life Insurance Pricing[J]. https://ssrn. com/abstract=2870308,2018.