DBSCAN
DBSCAN (Density-Based Spatial Clustering of Applications with Noise) · Also known as: DBSCAN Kümeleme, density-based clustering, density-based spatial clustering
DBSCAN is a density-based clustering algorithm, introduced by Ester, Kriegel, Sander and Xu in 1996, that groups together points lying in dense regions and flags points in sparse regions as noise. It is effective on noisy data and on clusters of irregular, non-spherical shapes.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
+15 more
When to use it
Use DBSCAN for exploratory clustering of continuous data when clusters may be irregularly shaped, when the number of clusters is unknown, and when the data contain noise or outliers you want to isolate rather than absorb. It assumes you can tune epsilon and min_samples (for example via a k-NN graph) and that cluster density is reasonably homogeneous. At least about 30 observations are recommended; below that, density estimates become unreliable.
Strengths & limitations
- Finds clusters of arbitrary, non-spherical shape rather than only round blobs.
- Identifies and isolates noise and outliers instead of forcing every point into a cluster.
- Does not require the number of clusters to be specified in advance.
- Effective on noisy spatial data where centroid-based methods struggle.
- Results are sensitive to the choice of epsilon and min_samples.
- A single epsilon struggles when clusters have widely varying densities; HDBSCAN or K-Means may be preferable.
- On small samples (below about 30) density estimates are unreliable and parameter tuning becomes meaningless.
Frequently asked
How do I choose epsilon and min_samples?
A common approach is to plot the distances to each point's k-th nearest neighbour and look for the 'elbow' in that curve to set epsilon; min_samples is then chosen as a small integer reflecting how many neighbours define a dense region.
What happens to points that are not in any cluster?
Points in sparse regions are labelled as noise and assigned to no cluster. This is a feature, not a flaw: it lets DBSCAN isolate outliers instead of distorting clusters by absorbing them.
When should I prefer K-Means or HDBSCAN over DBSCAN?
If clusters have widely varying densities, a single epsilon is inadequate and HDBSCAN or K-Means may work better. On very small samples, simpler methods such as K-Means are also safer because density estimates become unreliable.
Does DBSCAN need the number of clusters in advance?
No. Unlike K-Means it discovers the number of clusters from the data's density structure, controlled only by epsilon and min_samples.
Sources
- Ester, M., Kriegel, H.-P., Sander, J. & Xu, X. (1996). A Density-Based Algorithm for Discovering Clusters in Large Spatial Databases with Noise. Proceedings of the 2nd KDD, 226–231. link ↗
How to cite this page
ScholarGate. (2026, June 1). DBSCAN (Density-Based Spatial Clustering of Applications with Noise). ScholarGate. https://scholargate.app/en/machine-learning/dbscan
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Hierarchical ClusteringMachine learning↔ compare
- Random ForestMachine learning↔ compare
- Support Vector MachineMachine learning↔ compare