RegressionSample
k-nearest neighbours regression
Predicts by averaging the targets of the k closest training points.
Synthetic data, generated and trained in your browser.
Try this
- 01Set k = 1: the curve becomes a staircase that passes through every point.
- 02Raise k to 20: the curve flattens and misses the peaks of the wave.
- 03On the Outliers data, compare uniform and distance weighting near the outliers.
How it works
- 1Store the training data; there is no training step.
- 2For a new x, find the k points with the closest inputs.
- 3Average their targets, optionally weighting closer points more.
What the controls teach
- Neighbours k
- How many neighbours vote. Small k follows every point; large k smooths towards the mean.
- Weighting
- Uniform treats all k neighbours equally; by distance lets the closest ones dominate.
Use it when
- Local patterns with no simple formula
- Small, low-dimensional data
Avoid it when
- High-dimensional inputs
- Large datasets, where every prediction scans all points