Association Analysis For Fraudulent Claims In Auto Insurance Market

Insurance fraud is a common phenomenon in the insurance industry.Among multiple types of insurance,automobile insurance is one of the high-risk areas of insurance fraud.At present,along with the popularization of the insurance market and the improvement of people's living standards,the number of domestic automobile ownership has been increasing year by year,and the automobile insurance business develops rapidly.At the same time,due to the imperfection of domestic insurance system and the relative lag of the practice of insurance fraud management,the problem of automobile insurance fraud has become increasingly prominent,which has become a huge obstacle to the healthy development of insurance industry.In this context,automobile insurance fraud identification has become a very important topic.On the basis of the existing research,this thesis classifies 58429 automobile insurance claims samples in 2016 from a large domestic insurance company into three categories: black samples(confirming fraud claims),white samples(confirming honest claims category)and grey samples(indistinguishable claims category).Then the Apriori algorithm and the FP-growth algorithm are used to make association analysis of the black sample set,and the thirteen frequent itemsets acquired are verified by the white sample set,the result turns out to show that these frequent itemsets are all valid.In this way,the assosiation rules between each frequent itemsets are obtained.It is proved that there are significant association rules under each1-frequent itemsets,and the results are used to identify fraudulent claims.Using the results to identify fraud claims in grey samples,when {20,22} appears,8498 frauds can be identified,accounting for 34.26% of the grey samples,and when {10,19,20}appears,6325 can be identified.When {19,20,5} appeared,there were 3621 identifiable frauds,or 14.60 percent of the grey samples.Finally,apply these frequent itemsets with significant discrimination to predict the fraud rate of the grey sample set and identify the sample of claims that are in larger suspect.The association analysis shows that the predicted results of the fraud rate of the grey sample set depend on thechange of support degree of frequent itemsets.The higher the support,the more reliable the results are.Therefore,this thesis value the frequent itemsets of the support degree and discrimination degree of 0.80 and 0.95 respectively,at this time the predicted results of the fraud rate of the grey sample set is 0.4598 and the fraud rate of all claim samples is 0.3752.This result is close to the estimation of the fraud rate given by existing research.Due to the lack of domestic research on insurance fraud identification,the research on this problem using data mining technology has just begun.Therefore,the work of this thesis is still at the exploratory stage,and its conclusions need to be verified further.However,it can be affirmed that it will be one of the important directions of academic research and management practice of insurance fraud that the association analysis is used to find the characteristics of fraudulent behavior and then identify the fraud claims.