DBSCAN groups points by local density instead of distance to a centroid. A point is a core point if at least min_samples points (itself included) fall within radius ε. Clusters grow by chaining together core points whose ε-discs overlap; points inside a core point's disc but not core themselves become border points, and everything else is noise.
- eps — too small and everything is noise; too large and separate clusters merge.
- min_samples — higher values demand denser regions, producing more noise.
- k-distance plot — sort every point's distance to its k-th nearest neighbor; the bend ("elbow") in that curve is a strong starting guess for ε.