Handling class imbalance in multilabel classification using ULMFit

I am trying to develop a tone classifier that could infer multiple tones from a given email. Most of the labelled examples are “neutral” whereas the other ones are “anger”, “disappointment” etc…

Is there a way to address this type of class imbalance. I have looked up libraries such as imblearn but somehow I couldn’t find a suitable solution.

Could anybody suggest a way?


This thread would give an idea on handling class imbalances in multilabel setting

1 Like