Extreme events: blending PCA and ICA [19]

Go to: Summary | Previous | Next   
Bullet points include: PCA offers magnitude, ICA offers meaningfulness. Ideally we would like the best of both. But one focuses on variance, the other on (e.g.) kurtosis? Fortunately, it is possible to blend the two, e.g. by Recasting PCA along the lines of projection pursuit (with an importance criterion involving maximising contribution to variance), and then Choosing a different importance criterion that blends together variance and (e.g.) kurtosis

Contents | Prev | Next | Library

Desktop view | Switch to Mobile