Research On Automobile Insurance Fraud Recognition Based On Intelligent Risk Control

As an important economic infrastructure,insurance provides important guarantees for all aspects of people’s lives and social stability.At the same time,insurance fraud brings huge losses to the insurance market every year and affects the normal economic order.Fraud in automobile insurance is particularly prominent.With the increasing number of automobiles in our country,automobile insurance fraud has become a more serious problem.Aiming at the problems that the large amount of data and the rapid changes in the current automobile insurance anti-fraud task,which make it difficult for auto insurance companies to deal with it,based on the idea of intelligent risk control,this paper designs an automobile insurance fraud recognition framework.Based on the modified stacking model,an automobile insurance fraud identification model is established.Firstly,this paper introduces the background of automobile insurance fraud,and systematically expounds the solution to the problem of anti-fraud of automobile insurance at home and abroad.Then this paper analyzes the status,existing measures and remaining problems of automobile insurance anti-fraud problems,and proposes the problems to be solved in this paper.By analyzing the deficiencies of existing auto insurance anti-fraud problems,an auto insurance fraud identification framework based on intelligent risk control is built.The identification framework is consisted of three modules,they are the auto insurance big data platform,the auto insurance fraud identification model and the auto insurance anti-fraud decision engine.It supports the storage,cleaning,aggregation,feature monitoring and derivation of massive data,the training of intelligent models,and real-time decision-making,model monitoring and iteration.Finally,the superiority of automobile insurance fraud identification framework based on intelligence risk control is analyzed.Secondly,this paper constructs the stacking model,improves it,and conducts empirical research on two public datasets.First,a series of exploratory analyses were carried out on the data.Through the pre-screening of the models,five sub-models in the first layer were determined.These models were used to rank the importance of the models themselves,resulting in each model.The parameters of each sub-model are tuned using Bayesian parameter optimization.And by improving the output of the sub-model,combined with the isolated forest,the Stacking model is optimized.Then,this paper evaluates the model.First,the AUC of the Stacking model reaches0.89 and 0.99,respectively,and the KS exceeds 0.7,which is better than the performance of the sub-model.Then compared the model effect before and after optimization,the AUC value of the optimized model increased by 0.045,and the accuracy and recall rate also improved to a certain extent.The PSI of the final optimized model is 0.07,which shows strong stability.The results show that the model constructed in this paper can effectively lift the accuracy and stability of automobile insurance fraud identification,which has strong application value.Finally,this paper analyzes the model results based on the actual business situation,and proposes a threshold determination method for the automobile insurance fraud identification model and a strategy for classifying the risk level of claims.At the same time,the influence of important features on auto insurance fraud is analyzed,which provides a reference and explanation basis for the decision-making of automobile insurance fraud identification.The automobile insurance fraud identification framework based on intelligent risk control proposed in this paper takes into account the problems and optimization methods in actual business,which has strong applicability to the problem of automobile insurance fraud identification.The established optimized stacking model combines the advantages of the supervised model and the unsupervised model also has certain reference significance in theoretical research.