In DBSCAN, we may get some labels as noise. How to calculate NMI in that case? For example, if I use ground truth and predicted labels (keeping noise as -1) we got ARI=0.8 and NMI=0.75. Result seems quite good, but no. of predicted noises are pretty high.
Can anyone guide me?
Clustering is an unsupervised learning technique used to group data based on similar characteristics when no pre-specified group labels exist. This technique is used for statistical data analysis across many fields.https://medium.com/@tarammullin/dbscan-2788cfce9389
Thank you for your reply @Shafagat Mahmudova.
Silhoutte index only provides the quality of clusters. But it can not suggest comparative analysis w.r.t the available ground truth. Also, it is infeasible for very large dataset.
DBSCAN is a density based clustering algorithm, There are density based evaluation criteria that can be used such CDbw (Composing density between and within clusters) and DBCV (Density based clustering validation).
here is a good explanation in this blog
https://towardsdatascience.com/explaining-dbscan-clustering-18eaf5c83b31
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs