DBSCAN Density Based Clustering

DBSCAN (density-based spatial clustering of applications with noise) is a clustering algorithm that groups points lying in dense regions and labels points in sparse regions as noise.

DBSCAN은 밀집된 영역에 있는 점들을 군집으로 묶고 희소한 영역의 점들을 잡음(noise)으로 분류하는 밀도 기반 군집화 알고리즘이다.

How do I find clusters of arbitrary shape without choosing the number of clusters in advance?

군집 수를 미리 정하지 않고 임의의 모양을 가진 군집을 어떻게 찾나요?

DBSCAN forms clusters from points that have at least a minimum number of neighbors within a radius epsilon, and it labels isolated points as noise. It fits data with irregularly shaped clusters and outliers, where the number of clusters is unknown.

When to use DBSCAN Density Based Clustering

• Clusters with non-spherical or irregular shapes
• Data containing outliers that should be flagged as noise
• Spatial point data such as coordinates of events or locations

When to use something else

• Clusters have very different densities: use HDBSCAN or OPTICS
• Roughly spherical clusters and a known number of clusters: use k-means clustering

Assumptions and requirements

• Meaningful distance metric for the features
• Features scaled to comparable ranges
• Clusters of broadly similar density
• Sensible choice of epsilon and minimum points (e.g., from a k-distance plot)

What the analysis reports

• Cluster labels including a noise label
• Number of clusters and noise points
• k-distance plot for choosing epsilon
• Silhouette score (excluding noise)
• Cluster scatter plot

How to report it

DBSCAN (ε = [value], minPts = [value]) identified [k] clusters and classified [n] observations ([value]%) as noise.

Related methods

• K-Means Clustering
• HDBSCAN
• Hierarchical Clustering

Run DBSCAN Density Based Clustering on Quorant