Accuracy & evaluation

MAPE

Also called mean absolute percentage error

MAPE averages |error / actual| across periods. It is popular, easy to quote, and badly behaved when actuals are near zero or when small items should not outvote large ones.

People like MAPE because '12%' sounds like a grade. Then a SKU sells one unit, the model says two, and that row contributes 100% by itself. Buyers do not feel pain that way.

sMAPE and WAPE exist because MAPE kept embarrassing production teams. If a vendor only reports MAPE, ask what they did with zeros.

There are still series where MAPE is fine: strictly positive, similar scale, no intermittent zeros. Revenue at company-week can be one of them. SKU-day grocery is not.