Text classification with ML

I am currently working on my very first ML project at my new work. Task is to classify incoming emails to 4 separate mailbox by ML model to reduce costs. My model that I have succesfully trained is reaching almost 90% probability on imbalanced dataset.

Now to the question, how to force my model to only classify such an email if predicted accurancy is at least 95%. Rest will go to the 5th mailbox for manual classification provided by human.

Any idea? Thank you a lot.

Michal

I expect that my model will only classify such emails that could be predicted with at least 95 % accurancy.