Classical ML, validation, metrics, interpretation, debugging · insight-1179
Handle DBSCAN's noise label as an explicit population in downstream tables, not as a cluster or a discard
- Evidence type
- documented fact
- Difficulty
- medium
- Roles
- Data scientist, Data analyst
- Industries
- retail, manufacturing, logistics
Summary: situation, action, reason, boundary
Situation
DBSCAN labels are joined back to a customer or sensor table and label -1 is used in group-bys, one-hot encoded as cluster -1, or dropped.
Action
Report the noise share for every eps and min_samples tried, exclude noise from cluster profiling, and give it its own treatment: manual review, assignment to the nearest cluster only under a distance cap, or a separate rule.
Reason
A sample that is neither a core point nor within eps of one is labelled noise, so the label records a failure to fit any cluster, not a group.
Boundary
Density-based segmentation and anomaly screening on mixed-density data.
Symptom or risk
Noise points have no shared structure, so a profile of cluster -1 is meaningless, while dropping them hides rare entities; small changes in eps move large shares in and out of noise.
Exceptions
When noise is the output you want, such as an outlier shortlist.
Trade-offs
A separate noise workflow costs reviewer time.
Pitfalls
Tuning eps until the noise share looks small without checking that clusters are still separable.
Lessons and concepts
Claims and their sources
In DBSCAN a sample that is not a core sample and is at least eps from any core sample is an outlier; if eps is too small most data are not clustered and are labelled -1 (noise), if too large clusters merge.