I want to know how to extract the most relevant variables of a model using ICA.
When using PCA I can just verify the eigenvalues and select the new variables based on its variance. If I have 100 variables, but only 60 have relevant variances I can ignore the other 40. How does this procedure work in ICA ?
PCA and ICA have quite different objectives.
The main goal of PCA is to obtain the most important characteristics from data. However, typically, ICA is not used for reducing dimensionality but for separating superimposed signals (i.e., the goal of ICA is to find new components (new space) that are mutually independent in complete statistical sense.
In sum, main difference between PCA and ICA is that the former looks for the best orthonormal basis to minimize the variance losses from the data set, while the latter looks into the main components of the dataset to expand from there such basis. That means that orthogonality is not a requirement anymore.
HTH.
Dr. Samer Sarsam
Dear Victor,
if i understand right you are interested in how to select the most appropiate components after applying ICA. If yes, this requires in many cases a problem specific solution. You might have a look for "permutation indeterminacy" in the context of ICA.
Greetings, Sebastian
