Ellipsoidal Covariance Matrix

250 Views Asked by At

What does it mean when you say "Covariance matrix are highly ellipsoidal"? I am reading Regularized Discriminant Analysis" by Jerome H. Friedman 1989 and it uses this term very often "Covariance matrices are all equal and Highly ellipsoidal "?

1

There are 1 best solutions below

0
On BEST ANSWER

The covariance matrix of a set of variables is always symmetric by definition, which means that it can be diagonalized by an orthogonal matrix.

That being said, we can visualize the matrix as an ellipsoid whose axes have sizes given by it eigenvalues, and that has been rotated by the corresponding orthogonal matrix.

In this context the term ellipsoidal is being used in opposition to spherical. Being highly ellipsoidal means that the eigenvalues have very different magnitudes.