Mean Absolute Percentage Error (MAPE)
Mean Absolute Percentage Error · Also known as: MAPE, mean absolute percentage deviation
Mean Absolute Percentage Error measures prediction accuracy as a percentage relative to actual values, expressing errors in units that are scale-independent and interpretable across datasets. Formalized by J. Scott Armstrong in 1985, MAPE is widely used in forecasting, supply chain, and business analytics where results must be communicated as percentage accuracy.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use MAPE when you need a scale-independent metric for comparing models across different datasets or products. It is ideal for business contexts where stakeholders think in terms of percentage accuracy. However, MAPE has a critical flaw: it is undefined when actual values are zero and behaves poorly when actual values are close to zero. Avoid MAPE if your data contains zeros or very small values.
Strengths & limitations
- Scale-independent: directly comparable across datasets with different scales
- Interpretable: expressed as percentage, easy to communicate to stakeholders
- Penalizes both absolute and relative errors appropriately
- Widely used in industry, familiar to business analysts
- Undefined when actual values are zero; problematic with zero-containing or intermittent data
- Asymmetric: a 50% under-prediction is penalized differently than a 50% over-prediction
- Biased: favors under-predictions over over-predictions when actual values vary
- Can be misleading with mixed positive and negative actual values
Frequently asked
What do I do if my data contains zeros?
MAPE is undefined when actual values are zero. Consider alternatives: symmetric MAPE, mean absolute scaled error (MASE), or relative absolute error. Or exclude zero values and report accuracy on non-zero subset separately.
Why is MAPE asymmetric?
When actual = 100 and predicted = 150, error is 50%. When actual = 100 and predicted = 50, error is 50%. But MAPE gives different values. A 50% under-prediction (predicted = 50) has 50% error, while a 50% over-prediction (predicted = 150) has 50% error only if actual = 100. This asymmetry is inherent to percentage-based metrics.
How does MAPE compare to RMSE?
MAPE is scale-independent (percentage), while RMSE is scale-dependent (original units). RMSE is easier to optimize; MAPE is easier to interpret in business contexts. Neither is universally better; use both.
When should I use symmetric MAPE instead?
Symmetric MAPE addresses asymmetry by using the average of actual and predicted in the denominator. Use it if you want percentage errors that are symmetric and do not penalize over-predictions more heavily than under-predictions.
Is a MAPE of 10% good?
It depends on the domain. In demand forecasting, 10% MAPE is often considered excellent. In stock forecasting, it might be poor. Always benchmark against domain standards and baseline models.
Sources
- Armstrong, J. S. (1985). Long-range forecasting: from crystal ball to computer (2nd ed.). New York: John Wiley & Sons. ISBN: 978-0471082010
- Hyndman, R. J., & Koehler, A. B. (2006). Another look at measures of forecast accuracy. International Journal of Forecasting, 22(4), 679-688. DOI: 10.1016/j.ijforecast.2006.03.001 ↗
- Kim, S., & Kim, H. (2016). A new metric of absolute percentage error for intermittent demand forecasts. International Journal of Forecasting, 32(3), 669-679. DOI: 10.1016/j.ijforecast.2015.12.003 ↗
How to cite this page
ScholarGate. (2026, June 3). Mean Absolute Percentage Error. ScholarGate. https://scholargate.app/en/model-evaluation/mean-absolute-percentage-error
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Mean Absolute ErrorModel Evaluation↔ compare
- Mean Absolute Scaled ErrorModel Evaluation↔ compare
- Root Mean Squared ErrorModel Evaluation↔ compare
- Symmetric MAPEModel Evaluation↔ compare