ClusteringSample
DBSCAN
Grows clusters from dense regions and labels isolated points as noise.
Synthetic data, generated and trained in your browser.
Try this
- 01On Moons, DBSCAN separates the two shapes that k-means cannot. Nudge ε by 0.01 either way: they fragment or merge.
- 02On Rings, shrink ε until the rings break into fragments and noise.
- 03On Uneven blobs, lower ε below 0.05: the tight cluster holds together while the loose one breaks apart.
How it works
- 1A point with at least minPts neighbours within ε is a core point.
- 2Core points within ε of each other join the same cluster.
- 3Points near a core point but not dense themselves are border points.
- 4Everything else is noise.
What the controls teach
- Radius ε
- The neighbourhood radius. Too small and everything is noise; too large and clusters merge.
- Min points
- How many neighbours make a region dense.
Use it when
- Arbitrary shapes
- An unknown number of clusters, with noise
Avoid it when
- Clusters with very different densities
- High dimensions