The Extraction And Analysis Of Canonical Correlation Features From Relatedness Data

Canonical correlation analysis(CCA)is a multivariate statistical analysis method to study the linear correlation between two random variables.It has many applications in feature extraction,but there is little discussion on the canonical correlation feature visualization.Based on this,this paper improves canonical correlation analysis method,and puts forward a new method—canonical correlation analysis based on Newton cooling model,and applies it to the canonical correlation problems of multiple groups of relatedness data,and realizes the visual extraction of canonical correlation features.The main ideas of canonical correlation analysis method based on the Newton cooling model are as follows: At first,the sample set in the same mode is divided into N classes of sample spaces according to a certain category,so that the data category information can be fully utilized.Then pair these N types of sample spaces in pairs to obtain N(N-1)/2 pairs of sample spaces,and perform canonical correlation analysis on each paired sample space(a total of N(N-1)/2 times canonical correlation analysis),and then get the corresponding canonical variable coefficient vector.Then the Newton cooling method is used to fuse the obtained first d pairs the absolute value of the canonical variable coefficient to obtain a new pair of canonical variable coefficient vectors.Therefore,N(N-1)/2 pairs of sample spaces will get a new pair of fused new Coefficient vector of canonical variables.Finally,under the same coordinate system,draw scatter diagram of the canonical variable coefficients corresponding to the N(N-1)/2 sample features according to the sample features,and observe the dispersion of the scatter diagram of the canonical variable coefficients.In this paper,the ratio of the maximum distance and the minimum distance and average distance between scatter points is used as an evaluation index of the degree of dispersion of the features scatter plot.A larger index value indicates a greater degree of scatter dispersion,indicating that the feature is more canonical correlation.Therefore,we can intuitively judge the canonical correlation of features through scatter diagram of canonical variable coefficients,so as to realize the visual extraction of canonical correlation features.In this paper,through experiments on three measured data sets(coronary heart disease patient data,autism patient data,and white wine quality data),we found that canonical correlation analysis method based on Newtonian cooling model extracts canonical features of multiple sets of correlation data.The performance is very superior,successfully extracting the canonical features of coronary heart disease,autism,and white wine quality,which is of great help to the clinical treatment of patients with coronary heart disease and autism,and is of great significance to the identification of white wine quality.The experimental results confirm the superior performance of canonical correlation analysis method based on the Newton cooling model proposed in this paper,and also show that the effect of extracting canonical features is importantly related to the size of the class number N of the sample space and the degree of correlation of the sample space,that is,the larger the number of sample classes and the correlation coefficient of the sample space,the better the effect of feature extraction.In addition,the Newton cooling method proposed in this paper integrates canonical variable coefficients not only allows us to make full use of the information between multiple pairs of canonical variable coefficients,but also show the characteristics of the extracted canonical correlation features in the relatedness data through the visualization method,thus enriching the application space of canonical correlation analysis in feature extraction.