Generalized Linear Model Of Diagnosis Method Based On The Data Deletion

In statistical diagnosis, the recognition of outlier and processing is a very important task, it’s good or bad serious impact on the whole diagnosis process. The diagnosis of outliers in biology, medicine, education, sociology and economy and so on have very important significa-nce. This paper mainly studies the properties of the leverage value of the generalized linear models and the generalized linear models based on the data deletion of the diagnostic method, the key research through the data deleted model, Cook distance, W-K statistics the three meth-ods to distinguish a point is whether for outlier or strong influence point.Based on the linear model, and summarized some excellent properties of the generalized least squares estimate, given some properties of the residual in the generalized linear model. Proof that the projection matrix of the generalized linear model can be broken down into two or more the projection matrix, and at the same time, according to the properties of the projection matrix concluded that some properties of the leverage value in the generalized linear models, use the leverage value to discuss the influence of the observed value of a point for the Fitted value.Based on the basic diagnostic thought of the data deleted model, according to the theorem proving concludes that the generalized linear models deleted a data points before and after the relationship between the estimators, and at the same time, through the discussion to delete this data point to the influence of the model discrimination this data point is for outlier or strong influence points. Finally through the Cook distance and W-K statistics discussed to delete a data before and after,the relationship between some statistics, the order to identify whether this point is outliers or strong influence point.