Research On Short-term Collective Settlement Based On Zero-inflated Model

Nowadays, the enumeration data has been applied frequently in every aspect of temporary society, such as economics, insurance, biology, medical science, agriculture and Internet.The most common way for processing such data is utilizing classical discrete distribution which includes Poisson distribution, the binomial distribution and negative binomial distribution to establish corresponding mathematic model.We find that in practice, however, mass data with value of zero exists along with the increasing of data. To illustrate, we can conclude that there are a lot of non-smoking people when we investigate the amount people are smoking in one day where there is much zero. We can also find that there are a lot of people who do not file a claim when we look into the claim frequency and claims in insurance claim research where there is much zero existing.Likewise, when we do some research on the side effect of certain medicine, there are also many situation without side effects which means zero.The amount of zero in the condition above run circles round the amount of zero generated from Poisson distribution, binomial distribution, negative binomial distribution and some other distribution. It is hard to avoid the result of inaccuracy when we still handle certain problems with common distributions.This phenomenon is called “zero-inflated”.To solve this problem, Lambert proposed ZIP model in the early ninety’s to establish zero count blended with nonzero count probability distribution.In this thesis, we will first discuss the background of zero-inflated model generation, the research findings and three types of risk models.Moreover, we will also discuss how to apply the zero-inflated model into different risk models compared with primary fitting degree which will offer new methods to deal certain risk models containing an excess of zero count.