Research And Application Of Bayesian Models With GEV

Generalized extreme value(GEV)distribution,which is the asymptotic distribution of extreme variable,has advantages of less parameters,high flexibility and with the change of shape parameter or scale parameter,the symmetry and skewness of the distribution can be changed,correspondingly.Those advantages of GEV distribution can be used to make up for some of the shortcomings of some models.On the other hand,the advantage of Bayesian parameter estimation lies in the efficient use of prior information and many algorithms available in parameter estimation of complex models,like Approximate Bayesian Computation(ABC),MH and so on.In this paper,two Bayesian models combined with GEV distribution were considered.One is binary regression model linked by GEV.We propose the distributed Bayesian method to overcome the low efficiency for massive data processing and apply to analyze the massive game data to have the prediction error less than 3.5%.The other is GEV based Bayesian hierarchical model.The advantages and several variants of this model were studied.Finally we apply this model to study the human behavior with the massive phone calling data.