Machine Learning

DBSCAN

Density-based clustering that needs no k. It grows clusters wherever points are packed tightly, follows whatever shape they take, and labels the sparse leftovers as noise. Step through the expansion, or draw your own dense blobs and watch it flood-fill them.

Random data

Neighborhood ε 42

Min points 5

Clusters found—

Noise points—

Click the canvas to add points. Restart, then Step, to watch clusters flood outward from dense cores.

What to observe

Hit Restart then Step. DBSCAN picks a dense point andfloods outward through neighbors within ε, so a cluster takes on whatever shape the data has, even a long crescent.
Grey points are noise: they never gather enough ε-neighbors to join anyone. Built-in outlier detection, for free.
The two knobs fight each other. Too small an ε (or too highmin points) shatters everything into noise; too large and separate clusters merge. The right density sits in between.

Shortcuts: space run/pause · s step · r reset · f fullscreen