I am working on an extremely imbalanced dataset to build a classification model. The number of classes is 53 classes. I use early stopping on the validation loss to prevent the model from overfitting.
I've noticed that some people online use early stopping but on the macro-F1. I'd like to know, which way is better (more correct?) or more convenient? especially when we work on an imbalanced dataset. I think that both ways are good, but the loss is more reliable to look at.
I've found this thread Early stopping on validation loss or on accuracy? which doesn't answer my question. Accuracy is very bad in considering the minority classes.
