Introduced on day 15
Group k-fold
A cross-validation scheme in which all rows of one group (booking reference, customer, aircraft) fall in the same fold, so no group is on both sides of a split.
In detail
Use it when the model will score groups it has never seen. If returning groups' histories are legitimately available at prediction time, holding groups out is too pessimistic. It does not handle time: fold training sets can still contain periods later than the test rows. Stratifying by label does not provide this guarantee.