1

I am looking for informations about metrics for classification with 3 unbalanced classes. I have following numbers of samples in every class:

1

As you can see two classes are quite balanced and one is not (and I can't add more samples to this class). I was looking for methods for imbalanced data and I know I can add class weights or use metrics like F1-score or balanced accuracy.

My questions are:

  1. Which method is better?

  2. Are there other good metrics for imbalanced class besides balanced accuracy, precision, recall, F1-score and ROC?

  3. If I will decide to use e.g. balanced accuracy then should I make something more? Or can I use it just like accuracy for balanced classes.

  4. If I will decide to make class weights can I use metrics in the same way like with balanced classes (e.g. common accuracy)?

jared
  • 31
  • 1
    All of the metrics you note are highly problematical (ROC somewhat less so). Instead, use probabilistic classifications and evaluate them using proper scoring rules. See the proposed duplicate. – Stephan Kolassa May 01 '22 at 12:55

0 Answers0