hughes-peaking-phenomenon-non-monotonic-accuracy
IN premise — summaries/2026/08/24/wiki-Curse_of_dimensionality-chunk-2.md
Created 2026-08-24T17:11:08+00:00
The Hughes/peaking phenomenon (Hughes 1968; Zollanvari et al. 2020) states that with a fixed training sample size, classifier predictive power first increases with added dimensions, then degrades beyond a critical dimensionality—accuracy is non-monotonic in dimensionality.
Summary
Adding more features to a classifier only helps up to a point; past a certain number of dimensions, the model actually gets worse at predicting because the fixed amount of training data can't fill the growing feature space. This means there is a practical ceiling on how many features you should include, and blindly throwing more at the problem will eventually degrade accuracy rather than improve it.