The Optimal BMS Of Linear Exponential Distribution Family

In the practice of insurance two important factors considered were claim number and claim amount during the course of determining insurance premium. This paper only considers claim number model. Common experienced premium systems include BMS,NCD and credibility model.F'oisson distribution etc. are often described as claim nuto mber ,and these distribution belong to Linear Exponential distribution family .In the paper undoubted estimation of μ(θ) is first gotten,then using two parameters Linear Exponential distribution to replace Gamma distribution which is often used as structure function,building simple claim number model :Exponential distribution model,discussing the optimal BMS under different computing rulers. Finally a special model is given : Negative binomial model.Build discrimination function on observational swatches having been classified by means of fishr discrimination method,and classify new observational swatches under a certain discrimination ruler.The paper just bases on this method and introduces into ratio factor to adjust the proportion sizes of distance in groups and between groups in the model,then impove fisher discrimination model,namely change max = (C~TBC)/(C~TSC) into max = ρC~TBC - (1 -ρ)C~TSC ,finally a few examples validate discrimination efficiency to be enhanced.