Skip to content
Back to search results

DBSCAN

Introduced by
Ester, M., Kriegel, H.-P., Sander, J. & Xu, X.
Year
1996
Type
Density-based clustering algorithm
Font size

What does it do?

DBSCAN identifies connected dense regions and labels observations outside those regions as noise.

It does not require a preset number of clusters.

The intuition behind the method

Clusters grow through nearby core observations; sparse observations need not be forced into a group.

When to use it

Use when density connectivity reflects the research question, especially for irregular spatial groupings with noise. Select parameters in the units and geometry of the chosen metric.

Limitations and considerations

  • A shared radius can struggle with groups having different densities.
  • Results depend on the distance metric and density parameters.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Can identify irregular cluster shapes and leave noise unassigned.

Applications

  • Exploratory density grouping and investigation of sparse observations.

How it works

How it works

Choose a distance metric, a neighborhood radius and the minimum neighborhood count.

An observation meeting the density requirement is a core observation.

Connect core observations through their neighborhoods and include reachable border observations. Remaining observations are noise.

1
Formula
N_\varepsilon(p) = \{q : \mathrm{dist}(p,q) \le \varepsilon\}

Common pitfalls

  • Treating a universal minimum dataset size as an eligibility rule.
  • Confusing the minimum neighborhood count with a general sample-size requirement.
  • Ignoring feature scales or interpreting every noise label as a scientifically verified outlier.

Related methods

Sources

  1. 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.

    Open publication reference
    Copied.
  2. scikit-learn DBSCAN documentation

    Open publication reference
    Copied.