I would be thankful if you help me on this .
I have a data set of [450*40] , 450 corrersponds to samples and 40 corresponds to variables and need to select the maximum 15 out of 40 variables which have the most contribution in prediction model that further will be developed.
it is necessary to mention that the variables are not in same scales (not yet normalized) .
I came across ''Princomp'' function in matlab ,didn't get much from matlab user guid and have so many doubts.
1- Is princomp is a suitable function for my aim ?else what?
2- should I normalize my variables before applying ''Princomp'' or using ''princomp'' my variables get normalized automatically?
3- Does the COEFF output contain the variable with most contribution?
4-since my data sets are not labled , how can I identifiy which variables have most /least contribution ?
5-In code where can I set the number of variables that I need , in my case 15 ?
6-My dataset is in Excel format , any problem?
7-can apply kernel PCA (KPCA) instead of PCA , what is the difference?
Thank you so much in advance.
Awaiting your fast reply