Using pca for features selections

I have 13 features from 100 breast thermal images, to detect breast cancer, which are (means, standard deviation, correlation, contrast, energy, entropy, skeweness, homogeneity, variance, smoothness, KURTOSIS, RMS and IDM) and I want to use them to train Ann for classification (benign or malignant). How could I use pca to get the best features? Should I normalize the values before? And when I apply pca I get coeff and score, should I use the score as the input of my Ann or is there an equation between coeff and my features to use for my Ann? Sorry for my long question but pca is still confusing me!!!

Answer 1

[Greg Heath]

PCA (Principal Coordinate Analysis) is a very useful method for regression (it ranks linear combinations of the original variables)

HOWEVER

PLS (Principal Component Analysis) is a more useful method for classification (it ranks the original variables) !

I do not understand why it is not covered more extensively (if at all!) in the elementary statistics texts..

I have several posts in BOTH the NEWSGROUP and ANSWERS.

Hope this is helpful.

THANK YOU FOR FORMALLY ACCEPTING MY ANSWER!

GREG