In Python’s sklearn.metrics, I am using classification_report to help me interpret my machine learning model on an imbalanced dataset. For example, here is a arbitrary classification report:

Classification Report

When reporting a metric, for example precision, do we report the macro average (0.51) or weighted average (0.98)? I would assume the macro average, since this punishes the poor model performance on the positive class. Am I correct with this?

I do not believe this is a duplicate question with text, because I am basically asking whether it would be more useful (or used more in practice) to use macro or weighted average.

I have tried searching online about this. I didn’t see a definitive answer, but I believe that we use macro average.