Mean absolute percentage error: the average of |actual − forecast| / |actual|, as a percentage. It is easy to explain, but it is undefined when an actual value is 0, explodes near 0, and penalizes over-forecasts more than under-forecasts.
Card 1 of 259. Answer: Mean absolute percentage error: the average of |actual − forecast| / |actual|, as a percentage. It is easy to explain, but it is undefined when an actual value is 0, explodes near 0, and penalizes over-forecasts more than under-forecasts.
About · Quizzes · History of AI · To suggest new questions or report an error send me a dm.
MAPE expresses forecast error as a percentage of the actual value, so it is scale-free and reads easily: "we are off by 8% on average". That makes it popular in demand and sales forecasting, and also misleading whenever actual values can be small.
Example: actual sales are 2, 100 and 100 units, and the forecasts are 4, 90 and 110. The percentage errors are 100%, 10% and 10%, so MAPE is 40%, driven almost entirely by the 2-unit day. Weighted by volume, the forecasts are off by 22 of 202 units, about 11%, which is the WAPE.
The asymmetry is the interview point. For non-negative forecasts, an under-forecast can be at most 100% wrong, while an over-forecast has no limit, so a model tuned to minimize MAPE learns to forecast low. Alternatives: WAPE, which divides total error by total volume; sMAPE, which divides by the average of actual and forecast but has its own quirks; and MASE (Hyndman and Koehler, 2006), which scales error by that of a naive forecast.
