The precision, as it is defined here "http://mlwiki.org/index.php/Precision_and_Recall " is for just one cluster with one class. How to calculate the overall precision with the formula mentioned in the link above?
Are the examples below valid?
http://mlwiki.org/index.php/Precision_and_Recall
https://stats.stackexchange.com/questions/15158/precision-and-recall-for-clustering#comment388272_80194
https://stats.stackexchange.com/questions/15158/precision-and-recall-for-clustering
https://stats.stackexchange.com/questions/15158/precision-and-recall-for-clustering
The best definition and insights are given in this paper. In addition , Kappa score and F1 score. https://peerj.com/articles/cs-348/
Isn't precision just the same as precision for any truth table? And isn't recall just sensitivity for@ a truth table? see: https://en.wikipedia.org/wiki/Receiver_operating_characteristic#:~:text=A%20receiver%20operating%20characteristic%20curve,its%20discrimination%20threshold%20is%20varied.&text=ROC%20analysis%20is%20related%20in,analysis%20of%20diagnostic%20decision%20making.
As a precision and sensitivity are probabilities you could always use the geometric mean of the class specific results to determine the central tendency for the system, or you could sum the diagonal over the total sum of the table (but you won't get separate value for precision and sensitivity this way).
Thank you all for your answers. But what i really need to know is the relationship to pass from a set of (class i, cluster j) precision to the overall precision. I tried the average of the max's for each cluster and class but i didn't get the same results using the formula in this link:
https://www.researchgate.net/deref/https%3A%2F%2Fstats.stackexchange.com%2Fquestions%2F15158%2Fprecision-and-recall-for-clustering.
@Dan Harris the diagonal sum is used in classification, because in clustering clusters are not organized to fit classes (cluster i may have most datapoints of class j).
- Badges
- Science topic
- Similar topics
- Academic Journals