The Study Of Linear Discriminant Analysis Based On Semi-supervised Class Label

The high-dimensional data reduction is a process that the sample data is produced from a high dimensional space to a low dimensional space through linear mapping or nonlinear mapping.In this way,the data is made in the low dimensional space after no loss of original data information intrinsic geometric structure and topological structure.The high-dimensional data reduction can avoid the Redundant and irrelevant data and reduce the data size,which is applied to many fields,such as the classification and visualization.Based on the classical linear discriminant analysis method,this paper combines the semi-supervised class label into the linear discriminant analysis method,which puts forward a new method of a semi-supervised linear dimension reduction.Meanwhile,it applies to the incomplete data dimension reduction.The main study is as follows.1.This paper introduces several improved algorithms of linear discriminant analysis,then introduces several prediction methods of incomplete data and analyzes the advantages and disadvantages of these algorithms.2.The existing the Regularized Linear Discriminant Analysis(RLDA)only focuses on the local geometric structure of data,without considering the class label,which leads to an ineffective result in classification.To solve above problems,this paper combines the semi-supervised class label into the linear discriminant analysis method through constructing the same neighbor graph and heterogeneous neighbor graph and maximizing between-class scatter and minimize the within-class scatter,it proposes a new method of linear discriminant analysis based on the semi-supervised class label.This method overcomes the shortcoming of RLDA method in constructing the regular terms without considering the geometry structure of similar neighboring points and heterogeneous neighbor points.As a result,it is capable of maintaining classification.3.In the process of data acquisition and obtain,due to the influence of environmental factors and human factors,missing values always contained in the data.The common method of incomplete data dimension reduction is either ignoring the incomplete data only to complete data dimension reduction,or using the statistical method.Estimate the incomplete data,and then produce dimension reduction.Based on the semi-supervised class label of linear discriminant analysis method,this paper intends to make a prediction on the incomplete data and achieve the reduction of high-dimensional data at the same time.With this method,it improves the classification accuracy of incomplete data.