This program uses LDA and PCA to analyze features from weka arff file. The projection on PCA and LDA space visualizes the goodness of the features. If the features are good enough to be classified well they should have some kind of separation when projected on a 1 dimensional LDA or a 3 dimensional PCA space.
This MATLAB script assumes that the arff file has 2 classes named "Positive" and "Negative". However, it can be extended into any amount of class labels.
I have a question about Regu parameter that is optional in LDA procedure. I receive smaller classifiaction error (smaller about 10%) when I set it to 1. Why does the paraemter influence the output result so much ?
Thanks a lot !
07 Jun 2010
I've made the LDA space from 1D to 2D. Also included some sample arff files to read.