Need suggestion

Pover-T Tests: Predicting Poverty
1

Hi All,

I am doing K fold split on training dataset and getting below result when calculating the log loss of individual country:
A: ~ 0.30
B: ~ 0.19
C: ~ 0.02

But in final submission, the combined log loss is coming as high as 1.16. Can anyone please point out where I am possibly mistaken? Thank you in advance!

2

Are you submitting the predictions rather than probabilities?

3

@vijayabhaskar oh yes! indeed we are using probability correctly. yet somehow, log loss is coming very high.
any suggestion to overcome this issue?