I am working on binary classification problems, and using pycaret to try out different methods. All my problems are imbalanced, so one class is typically around 10% of the data.
When I use pycaret to pick a threshold which optimizes for say f1, I often get very low optimal thresholds like 0.18 or sometimes even as low as 0.1. I wanted to know if this is an inherent issue? Typically, I would expect thresholds close to or higher than 0.5 for a good model. Is this normal behavior?
I am already checking for overfitting by checking the validation and test set metrics to ensure they are similar ...
