Does anyone have a reference that can be cited where it is stated to retain the number of principal components needed to explain 90% of total variance?
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Even if you find such a publication I think it is a bad rule of thumb. It is better to look at the eigenvalues and look for this characteristic "drop" in eigenvalues. If your first three dimensions explain 35%, 32% and 22% of the variation it is not relevant to include a fourth that explains only 1% to get to the magical boundary of 90%.
