The Algorithm And Data Analysis About Sufficient Variable Selection Via Ball-Covariance

Variable selection,which is an important part of dimension reduction in Statistic,has been widly used in our daily life and data analysis.However,we are in the era of big data which means the number of data and variable is becoming higher and higher.Sometimes we can not use traditional variable selection methods any more.So try to use other methods or find some important variables which can be used in variable selection,is becoming a new subject needed to be researched.In our paper,we propose Ball-Covariance,Bcov,and Ball-Correlation,Bcor,to do variable selection.On the basis of previous studies of using DC and HSIC to do variable selection,we compare the features of these three kinds of independent measures and try to find the similarities between them.After that we think these three kinds of independent measures are similar.So we assume that we can try to do variable selection via Bcov methods.Based on the principle of Analogy,we think we can propose such a hypothesis.According to the previous studies about Bcov and our hypothesis,we design our data test.The results of our test show that using Bcov and Bcor to do variable selection is a feasible and correct method.In the end,we do data analysis to explain the practicality of our method.