Principal Component for Shape Analysis.

Hello guys,

I want to make a pca for my shape problem. My X matrix is ​​constituted n = rows (observation) and p = columns (variables). In particular, my matrix is ​​constituted by n = 15000 which are the coordinates of the geoemtrical points of the form and m = 10 are the pazinets. I run the following script in Matlab:

[PC,Mode,eigenvalues,tsquared,variace_perc,Mean_RR] = pca(RR);
Xcentered = Mode*PC';
shape_vector=mean(Xcentered,2)

I would like to be sure to proceed in the right way. From this analysis I get:

PC=matrix of 15000*9

Mode=matrix of 10*9.....

In this way PC represent the principal components for me, are just projection of my date onto the principal component.

Mode is that Matlab call "coeff"....but in this way, what does it represent?

Thank you very much!!!

Answer 1

[Bernhard Suhm]

Per documentation of pca, your "Mode" are " .. the representations of X in the principal component space. Rows of scorecvorrespond to observations, and columns correspond to components". So to get the loading of the 2nd component of the shape vector corresponding to your input signal, you'll want to grab the second column of your "Mode" variable.

[Fredic]

Thank you very much for yours answer.

I understood the meaning of the various parts of the pca. I have a doubt about my input matrix. I would like to insert two different types of patients, a healthy group and a group subject to a certain pathology. how can I, in such a structured way, make my system understand what is "normal" and what is not?