DBSCAN
Density clustering that can label points as noise.
When to use it
Use DBSCAN when you want clusters of dense points and you are willing to leave sparse points as noise, rather than forcing every row into k groups. A fixed k is cluster analysis.
Assumptions
Columns are numeric. Epsilon starts at 0.5, minimum samples at 5, standardize on. HTTP and chat dispatch both honour standardize. If more than 30% of rows are noise, the report warns. There is no k to choose.
Running it in Tensr
Options
Prop
Type
Reading the output
The two blobs, ε = 0.8, min_samples = 5, standardised. Dense cores should appear, with a modest noise share.
| Cluster | Size |
|---|---|
| 0 | 120 |
| 1 | 120 |
2 dense cluster(s), 0 noise points (0%). ε = 0.8 on two well-separated blobs after standardising should recover the cores and leave a modest noise share.
Reporting (APA 7)
DBSCAN on x and y, ε = 0.8, min_samples = 5, 2 cluster(s), 0 noise points.
Coming from SPSS
SPSS has no DBSCAN dialog. Tensr’s item is on the ML menu, Clustering → DBSCAN. The path string stored for this item matches that menu.
Related
A chosen number of groups is cluster analysis.