Skip to content
ML Academy

Introduced on day 16

MAPE (mean absolute percentage error)

The mean of absolute errors divided by the actual value, times 100; undefined when any actual is zero and unstable near zero.

In detail

MAPE = (100/n) x sum(|y - yhat| / |y|). Each row counts equally regardless of volume, so many small units can outvote a few large ones. It penalizes over-forecasts more than under-forecasts, so models tuned on it drift low. Report it as undefined when any actual is zero; consider WAPE or a scaled error instead.