Description
k-Nearest Neighbors is a non-parametric classifier / regressor that predicts using the labels of the k closest training points under a chosen distance metric. It is the canonical query engine for csi-fingerprinting and the simplest possible CSI-based localiser: at inference time, look up the most similar fingerprint in the radio map.
When it's used
- Fingerprinting-based indoor positioning
- Quick-and-dirty CSI classification baselines
- Distance-based open-set rejection prototypes
Limitations
- Inference cost scales with database size
- Sensitive to distance metric choice on high-D CSI vectors
- Storage cost grows with site coverage