Research And Improvement Of Fisher Discriminant Analysis Method

The discriminant analysis is an important branch in statistical study, it seriously affect the whole research process. The discriminant analysis always use together with the cluster analysis. Such as discriminant analysis request research object must be known, we can classifies unknown sample. And discriminant analysis fit this request, so that we can use discriminant analysis classify sample data, then according to the result determined the discriminate function in order to ensure the unknown sample belong to which category.This paper introduced some methods of discrimination function model, discrimination criterion and identifying steps. And compare these methods in the second chapter. This paper has given the basic thought, discrimination function, discrimination criterion of the distance discrimination method, Bayes discrimination and Fisher discrimination. Then introduced the determination method of the definition and critical value which are belong to Mahalanobis distance and Cumulative discriminate ability. Then checkout the Fisher discrimination whether affection and density function and posterior probability of Bayes discrimination.The innovation research of this paper is improve traditional Fisher discrimination model. Introduce three methods such as an extraction method based on Fisher norm, optimal norm concentrate method, Multiple discriminate of Fisher discrimination. Based on this raised a improved method, this means Introducing weighting factor. This method based on discrimination function which is under the primary function, then fractional line discriminate function became differential line discriminate function in order to structure poor value maximization model. At last, choose optimal p value so as to we can get the best discrimination function and enhance the discriminate efficiency.