Knowledge Maxxing← Back to topic library
ML MaxxingBeginner60 sec
Mean absolute error
Mean absolute error is the average size of a regression model's prediction mistakes, ignoring whether each mistake was too high or too low.
Prerequisite: Regression
What matters
Because it treats every unit of error equally, MAE gives an easily interpretable 'typical mistake size' in the same units as the predicted value.
Example
If a house-price model's predictions are off by $10,000, $20,000 and $30,000 on three houses, its mean absolute error is $20,000.
Common misconception
Mean absolute error does not penalize large errors more than small ones; each unit of error contributes equally, unlike mean squared error.
Recall question
Why does MAE treat a $50,000 error the same per-dollar as a $5,000 error?
Keywords and related concepts
mean absolute errorMAEregression metricmean squared error