DBSCAN Clustering

DBSCAN finds clusters as connected regions of high density: a point is a core point if at least minPts neighbours lie within radius eps, and clusters grow by expanding outward from core points through their neighbourhoods. Choose a preset — two moons, blobs with sprinkled noise, nested rings, or varying-density clusters — or click to add points, then set eps (a to-scale circle next to the slider shows the radius) and minPts and press play: the region-growing queue is animated point by point with the current frontier highlighted. Core points are drawn solid, border points hollow, and noise as gray crosses. Unlike k-means, DBSCAN needs no cluster count, follows arbitrary shapes such as moons and rings, and labels stragglers as noise — but the varying-density preset exposes its weakness: one global eps cannot fit dense and sparse clusters at once.

Runs 100% in your browser — models are trained and computed locally on your device.

Read the full guide to this tool

Notes

A core point has ≥ minPts points within eps; a border point is within eps of a core point but is not core itself; everything else is noise. Clusters are the connected components of core points.
Naive neighbour search makes this implementation O(n²); with a spatial index DBSCAN runs in O(n log n). Border points reachable from two clusters are assigned to whichever claims them first, so labels can depend on visit order.
A single global eps is DBSCAN’s main weakness: if one cluster is dense and another sparse, no eps separates both correctly — that is what OPTICS and HDBSCAN were invented to fix.
A rule of thumb: set minPts ≈ 2·dimensions and read eps off the knee of the sorted k-nearest-neighbour distance curve.
Runs 100% in your browser — models are trained and computed locally on your device.

DBSCAN Clustering

Notes

Related tools

Distribution Sketcher

PCA Toy

k-Means Clustering

Anomaly Detection Playground

Gaussian Mixture & EM

Hierarchical Clustering & Dendrogram