ZHANG Shu,TAO Hongcai.Research on Prediction Model of Auto Insurance Claim based on Improved DeepFM[J].Journal of Chengdu University of Information Technology,2021,36(03):311-315.[doi:10.16836/j.cnki.jcuit.2021.03.012]
基于改进DeepFM的车险索赔预测模型的研究
- Title:
- Research on Prediction Model of Auto Insurance Claim based on Improved DeepFM
- 文章编号:
- 2096-1618(2021)03-0311-05
- Keywords:
- auto insurance claim; feature interaction; DeepFM; attention mechanism
- 分类号:
- TP303
- 文献标志码:
- A
- 摘要:
- 广义线性模型因其简单且输出结果具有可解释性被广泛应用于车险索赔预测领域,但不能识别特征之间交互作用从而限制了模型的表现力。DeepFM使用因子分解机和深度神经网络分别捕捉低阶和高阶特征交互,在数据稀疏的实际场景取得了显著效果。在因子分解机的基础上引入域相关的权重,针对特征存在互相干扰的问题提出相应缓解策略,并将轻量级的视觉注意力机制作用于深度神经网络进一步提升模型的表现力。实验结果表明,提出的模型相比于基本的DeepFM模型取得了更好的风险分割效果。
- Abstract:
- Generalized linear model is widely used in the field of auto insurance claim prediction because of simplicity and interpretability. However, its expressiveness is limited because it can’t recognize the interaction between features. DeepFM uses Factorization Machine and Deep Neural Network to capture the interaction of low-order and high-order features respectively, and achieves remarkable results in the real scene with sparse data. This paper introduces the weight of domain correlation based on Factorization Machine, and proposes mitigation strategies to ease the problem of mutual interference between features. The lightweight visual attention mechanism is also applied to the Deep Neural Network to enhance the accuracy of the model. Experimental results show that the proposed model achieves better risk segmentation effect than the basic DeepFM model.
参考文献/References:
[1] 曾宇哲,吴嫒博,郑宏远.基于机器学习的车险索赔频率预测[J].统计与信息论坛,2019,34(5):69-78.
[2] 吴育文.广义线性模型在车险精算定价中的实证研究[J].内燃机与配件,2018,267(15):190-193.
[3] 孟生旺,李天博.基于机器学习算法的车险索赔概率与累积赔款预测[J].保险研究,2017(10):42-53.
[4] 张连增.回归树方法在车险索赔频率预测建模中的应用[J].保险研究,2018(1):101-111.
[5] 薛智雯.基于ARIMA-SVM的车险索赔次数预测[D].成都:西南财经大学,2018.
[6] 孟生旺.神经网络模型与车险索赔频率预测[J].统计研究,2012(3):22-26.
[7] 张连增.提升算法对传统车险索赔频率建模模型的改进——基于我国五省交强险保单数据[J].保险研究,2019(7):67-78.
[8] Lee S,Antonio K.Why High Dimensional Modeling in Actuarial Science[C].Proceedings of the IACA Colloquia,2015:75-79.
[9] Xiao J,Ye H,He X,et al.Attentional factorization machines:Learning the weight of feature interactions via attention networks[C].Proceedings of the 26th International Join Conference on Artificial Intelligence,2017:3119-3125.
[10] Rendle S.Factorization machines[C].Proceedings of the 2010 IEEE International Conference on Data Mining,2010:995-1000.
[11] Rendle S.Factorization Machines with libFM[J].ACM Transactions on Intelligent Systems and Technology,2012,3(3):1-22.
[12] 孙志军,薛磊,许阳明.深度学习研究综述[J].计算机应用研究,2012,29(8):6-10.
[13] Cheng H T,Kocl,Harmsen J,et al.Wide & deep learning for recommender systems[C].Proceedings of the 1st workshop on deep learning for recommender systems,2016:7-10.
[14] Guo H,Tang R,Ye Y,et al.DeepFM:A Factorization-Machine based Neural Network for CTR Prediction[C].Twenty-Sixth International Joint Conference on Artificial Intelligence,2017:1-8.
[15] Roy G,Navab N,Wachinger C.Concurrent Spatial and Channel Squeeze & Excitation in Fully Convolutional Networks[C].International conference on medical image computing and computer-assisted intervention,2018:421-429.
[16] 黄秋彧.个人信用风险评分的指标选择研究[J].新疆财经大学学报,2015(3):5-15.
备注/Memo
收稿日期:2020-03-28
基金项目:国家自然科学基金资助项目(61806170)