MAE (Mean Absolute Error)

Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary

The average absolute difference between predicted and actual values: a regression loss that weighs every error in proportion and resists outliers.

What is MAE (Mean Absolute Error)?

MAE (mean absolute error) is a regression loss that averages the absolute differences between a model’s predictions and the actual values. Every unit of error counts the same, so one extreme miss cannot dominate the score.

Its second virtue is readability: MAE stays in the target’s own units. An MAE of 2 minutes on arrival-time predictions means the model is off by 2 minutes on average, a number anyone can act on. For classification tasks, cross-entropy loss plays the role MAE and MSE play in regression.

How MAE Works

\[ MAE = \frac{1}{n} \sum_{i=1}^{n} |y_i - \hat{y}_i| \]

where \(y_i\) is the actual value, \(\hat{y}_i\) the prediction, and \(n\) the number of data points. Taking absolute values removes the sign of each error without amplifying its size.

That linearity is what makes MAE robust: an error of 100 costs exactly ten times an error of 10, not a hundred times. The trade-off is at the training stage, where the absolute value has a kink at zero and a constant-magnitude gradient, which some optimizers handle less gracefully than MSE’s smooth curve.

MAE vs MSE

The practical difference: MAE weighs every unit of error equally, while MSE punishes large errors far more than small ones – so MAE is robust to outliers and MSE is sensitive to them. Choose MAE when a few extreme cases should not dominate the measure, and MSE when large errors are disproportionately costly.

CriterionMAEMSE
Penalty on large errorsProportional, so outliers count onceSquared, so outliers dominate the total
Units of the resultSame units as the target (e.g. dollars)Squared units of the target (e.g. dollars²)
Gradient behaviorConstant magnitude, with a kink at zeroGrows with the error, smooth everywhere
Choose it whenOutliers should not dominate the scoreLarge errors are especially costly

Example of MAE

A ride-hailing app predicts trip durations. Across 1,000 trips, its predictions miss the true duration by 3 minutes on average, so the MAE is 3 minutes, stated in the same units riders think in.

Now one trip hits an accident and runs 90 minutes over the prediction. Under MAE, that trip adds 90 error-minutes to the total, raising the average only slightly. Under MSE it would add 8,100 squared minutes and swamp the other 999 trips.

If the goal is a fair picture of everyday accuracy in time series forecasting like this, MAE gives it. The one freak trip stays one data point, not the whole story.

Related AI terms: MSE · Cross-Entropy Loss · Regression Analysis · Time Series · Model Evaluation

Did you like the MAE (Mean Absolute Error) gist?

Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.

Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About

Read the Governor's Letter

Stay ahead with Governor's Letter, the newsletter delivering expert insights, AI updates, and curated knowledge directly to your inbox.

By subscribing to the Governor's Letter, you consent to receive emails from AI Guv.
We respect your privacy - read our Privacy Policy to learn how we protect your information.

Browse All AI Terms A–Z

Every term in the dictionary, in alphabetical order. Jump to a letter or scroll the full list.

A

B

C

D

E

F

G

H

I

J

K

L

M

N

O

P

Q

R

S

T

U

V

W

X

Y

Z