what is the COEFF,SCOR​E,latent,t​square when perform PCA

Hi,

I am new to use PCA, I have gone through, but still unable to capture what is the mere meaning(physical significance) of these terms:COEFF,SCORE,latent,tsquar (how they are important).

Many thanks in advance,

Answer 1

[KSSV]

I suggest you to read first what PCA does and how it is done. Then you will get clear picture. Also know that MATLAB documentation is vast and the required information is Crystal clear there.

Let x be your input.

coef-are the principal components aka eigenvectors of covariance matrix of x arranged in descending order.

score-it is the input x rotated to new basis of principal components.

latent-these are eigevalues of covariance matrix of x arranged in descending order.

PCA is used for dimensional reduction. Now instead of using the whole x , you can use certain columns of score for analysis.

[nkwt nkwt]

Thank you. It is what I have been looking for too. Wondering why the MATLAB doesn't describe these variables in relation to the eigen components?

I understand an approximation matrix of the original features x can be reconstructed using score * coeff' (using the selected high rank features). But, is there a way to refer to which features is mapped to which reconstructed Feature, as these are already sorted by pca?