Wednesday, February 21, 2018

SKLearn Metrics Precision, Recall, F-Score and Support

Precision:
  • If it's 1 it means all positives are correct
    • It may be that many other positives for this class are wrongly labelled as negative (and categorised in other classes), but that's not measured in this metric
  • The lower it is, it means more entries from other categories are being mis-classified here
Recall:
  • If it's 1 it means it managed to find all the positive samples
    • It may be that it wrongly classified other samples for other categories here, but that's not measured in this metric
  • The lower it is, it means more entries from this category are being mis-classified into other categories
F-Score: Some kind of mean between precision and recall

Support: Number of occurrence in each category

Reference: http://scikit-learn.org/stable/modules/generated/sklearn.metrics.precision_recall_fscore_support.html

No comments:

Post a Comment