Model incorrectly classifying intent confidence

Hi everyone, I have trained my rasa model with 4 different intents. I also specified a fallback threshold, so that if the confidence for the intents are below that threshold a fallback page will be displayed.

However, there is always 0.97+ confidence score for one of the intents, even if the input i gave is completely different from any of the training data. This is confusing for me, it makes sense that all the 4 intents should have a very low confidence score if they are given an input that is completely random and new.

Has anyone had this problem?

Hi @aaronlikesrasa, can you share your implementation?

And an example for this?