Multicollinearity

Features that are strongly correlated with each other, which makes the individual coefficients unstable and hard to interpret.

When to use

In PS 04, car age and kilometres driven are correlated: without regularisation they fight for credit and coefficients swing with small data changes, while Ridge Regression spreads the shared effect. PCA components are orthogonal, so PCA also removes it.

See Lec 4.

Appears in

No references yet.