I have watched a lot of videos on machine learning and in terms of F1 scores, all are different. One video says that an F1 score of .8 is bad, but another says an F1 score of .4 is excellent. What's up with this?

I ran my model with Random Forest algorithm and got a modest average of .85 after about 5 folds. After I used my undersampling approach, I had an F1 final score of about .92-.95 after 5 folds.

If you're wondering what it was about, it was basically random numbers which was previously credit card fraud data but was replaced because of sensitive information.

Basically, my question is, what range of F1 scores are good, and is my f1 score of .92-.95 good?

(An F1 score is the harmonic mean between the precision and recall of a dataset.)


  • $\begingroup$ Not very good, it's based on human performance (ignore number of examples) in same problem and problem sensitivity. $\endgroup$ – 4.Pi.n Jul 26 at 3:57
  • $\begingroup$ It would help if you explained what an F1 score is (and also, if your example is 'basically random numbers', why you'd expect them to be predictable). $\endgroup$ – Thomas Lumley Jul 26 at 4:52
  • $\begingroup$ Hi Thomas look at my edit. Also the dataset was from Kaggle and it had about 6000 upvotes and I think he made a proportional replacement for the classes. $\endgroup$ – Sriswaroop Koundinya Jul 26 at 15:06
  • $\begingroup$ It depends on the task. A F1 score of 0.8 might be disappointing or it might make you the richest person on planet (if you are predicting if a stock will gain/lose). Make sure that you evaluate on a 100% independent data set though. $\endgroup$ – Michael M Jul 26 at 16:07
  • $\begingroup$ Oh, thanks Michael! $\endgroup$ – Sriswaroop Koundinya Jul 26 at 16:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.