Regularized Bilateral Two-Dimensional Linear Discriminant Analysis

Linear discriminant analysis is an important classification method for areas such as climate classification,agricultural planning,medical research,face recognition,and credit risk management.There are a lot of redundant information and noise information in face recognition data,multivariate time series and other high-dimensional data.When the method is used for discriminating classification,destroys the original matrix structure of the data,and brings great difficulties to feature extraction.making it difficult to experiment.Compared with linear discriminant analysis,two-dimensional linear discriminant analysis directly extracts features from two-dimensional matrix of images,saves the matrix data's natural matrix structure,avoids the problem of dimension disaster to a certain extent,has high efficiency,and also significantly reduces the storage space and computing intensity of computers.At present,there have been a number of extension algorithms for discriminant analysis,but many of them still have a large characteristic dimension of the divergence matrix between classes and within classes.When traditional two-dimension linear discriminant analysis processes high-dimensional matrix data,the classification accuracy is unstable or difficult to distinguish when the distance between the inter-class mean and the global mean is close to the inter-class mean by the inter-class divergence matrix.Regularization technique for solving ill-posed problems of inverse has very good effect,in order to enhance the generalization ability of the model,and limit the complexity of the model,on the objective function to join prevent regularization item is a kind of model fitting increase the generalization ability of the commonly used methods,such as regression and regression in the regression is a very good effect of regularization.In order to find a better classification and discrimination algorithm,this paper aims at the defects of the feature dimension of the inter-class and intra-class divergence matrix with high and medium dimensions.In this article,through expanding regularization method to bilateral two-dimensional discriminant analysis,regular bilateral two-dimensional discriminant analysis is put forward,namely the bilateral two-dimensional discriminant analysis on the horizontal direction and vertical direction of the class in the process of feature extraction in divergence on the matrix combined with regularization item,respectively,to extract the characteristic information of the more understanding in order to achieve higher recognition rate of algorithm,can have better discriminant effect.In this paper,Five time series data verification methods are selected through experiments and compared with experiments to obtain the corresponding optimal dimension and classification error class,the optimal regularization parameter of the regularization method is determined by cross-validation method.and it is proved that the method is significantly improved in the reduction and classification accuracy.