Introduction
Look at a night-time satellite photo of a country: cities appear as bright, dense clusters of lights of all shapes — long coastal strips, round towns, sprawling metropolitan areas — separated by dark countryside, with a few isolated farmhouse lights here and there.
DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clusters data the same way: a cluster is a continuous region of high density. It:
- finds clusters of any shape (rings, moons, long strips);
- figures out how many clusters there are by itself;
- labels points in sparse areas as noise (outliers) instead of forcing them into a cluster.