Curse of Dimensionality

In many dimensions all points become almost equally distant from each other, so the nearest neighbour is hardly closer than the farthest. Distance-based methods degrade, models need far more data, and visualisation becomes impossible.

When to use

It explains why clustering on 50+ raw features is unreliable and why KNN and K-Means suffer. The lecture's fix is to reduce dimensions first with PCA, then cluster.

See DS Lec 5.

Appears in

No references yet.