I Know that the supervised method is evaluated in terms of precision, recall and f1 measure. Therefore, what evaluation criteria is used for the evaluation of unsupervised method? Can an unsupervised method be evaluated in terms of precision, recall and f1?
Hi Jibran,
To evaluate unsupervised learning schemes, you could rely on the log-likelyhood metric when using the density-based approach. Another option is converting the unsupervised learning into a supervised learning technique, where you can use its evaluation metrics such as ROC, F-measure, etc.
HTH.
Samer Sarsam, PhD.
I do not think it is possible. Even if you have labelled data for calculation and comparison, you will have a trouble of how to match clusters to the labeled classes. I have tried this somehow with no real breakthrough. These are some crude unpublished works that I have done on the subject. If you wish we can collaborate to continue it and see what can we do.
Alright, tell me then what is the evaluation criteria for unsupervised methods
- Badges
- Science topic