Just as the title imply, I am searching for an error evaluation metric with the following characteristics:
- able to handle cases where actual value is 0
- can evaluate different units/scales
- isn't disproportionately affected by outliers
My intention is to find a complementary evalution metric to sMAPE.