So I wrote an article using data from this competition. Read and share. ✨
https://saheedniyi.hashnode.dev/handling-imbalanced-datasets-tuning-classification-threshold
https://twitter.com/saheedniyi_02/status/1571807969897025539?t=6GHm5iQ3x0QaPIa8WNPfIA&s=19
Thanks, man! I was sure, that have to use 0.5 threshhold all the times :) Good to know now. One Question - I've tried to stratify train_test_split, but it didn't help, despite big data disbalance. Any idea - why?
Thanks so much! going to see if I can put that to work in my next submission