liveBook · Manning

Mean absolute percentage error

As explained in Machine Learning System Design

The average absolute prediction error expressed as a percentage of the corresponding actual values.

Mean absolute percentage error (MAPE) averages absolute forecast errors as percentages of their corresponding actual values. It looks intuitive for communicating forecast accuracy, but it becomes undefined whenever a future actual value is zero. section 5.3.1

Use MAPE cautiously when evaluating forecasts:

  • Reject it for count-data forecasting when zero-demand periods are possible, including supply-chain and inventory settings. section 5.3.1
  • Compare it with MAE, wMAPE, sMAPE, MSE, and relative-error measures rather than treating percentage error as universally appropriate. section 5.3.1
  • Remember that sMAPE avoids undefined values when only one side is zero, but assigns a contribution of 2 whenever the actual is zero. section 5.3.1

How Machine Learning System Design covers it

How deep it goes
Defines MAPE through its zero-denominator failure and contrasts alternatives, without deriving or implementing the metric.
Examples
Uses future count-data realizations, including zero-demand periods, to expose MAPE’s undefined result.
What you can do after
Evaluate forecast metrics by checking whether future actuals can be zero before selecting MAPE.
How it connects
Connects MAPE to MAE, wMAPE, sMAPE, MSE, and relative-error measures in forecast evaluation.
Cover of Machine Learning System Design
liveBookmanning.com
Read Machine Learning System Design on liveBook Get the book at manning.com

Also explained in

Compare how 3 books explain mean absolute percentage error →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Machine Learning System Design. The section links point to the parts of the book it draws on.