Skip to content
BENIBIX Learn

MAE vs RMSE

People think RMSE is just another way to write the average miss, like MAE. It is not. MAE counts each miss by its size. RMSE squares each miss first, so one big miss pulls it up far more. When the two are far apart, a few wild misses are hiding in the data.

AI📏 Evaluation & Benchmarks

Mean Absolute Error (MAE)

The average size of a model's misses when it guesses numbers. Lower is better.

Mean absolute error, or MAE, measures how far a model's guesses are from the true numbers. For each guess, you take the gap and drop any minus sign. Then you take the average of all those gaps. The result is in the same unit as what you predict, such as euros or degrees. An MAE of 2 degrees means the guesses are off by 2 degrees on average. Lower is better, and 0 means every guess was exact.

Example
Jack's model guesses each day's top temperature in London. Over 30 days, its guesses are off by 1.5 degrees on average, so its MAE is 1.5.
Formula
MAE = average of |true value - guess|
Why it matters
It is the easiest error score to explain. Anyone can read it as the usual miss, in real units.
Full entry
AI📏 Evaluation & Benchmarks

Root Mean Squared Error (RMSE)

An error score for number guesses that punishes big misses more. Lower is better.

Root mean squared error, or RMSE, scores a model's guesses on numbers. You take each gap between a guess and the true value and square it, so big gaps grow much bigger. The average of those squares is the mean squared error, or MSE. Then you take the square root to get back to real units. Because of the squaring, a few big misses push RMSE up a lot. It is always at least as big as MAE on the same guesses.

Example
Two weather models both have an MAE of 2 degrees. One misses by 10 degrees on a few days, so its RMSE is much higher. Emma picks the other one.
Formula
RMSE = square root of the average of (true value - guess)²
Why it matters
When one huge miss is worse than many small ones, RMSE shows it. Many models are trained to make the squared error small.
Full entry

In one line

Mean Absolute Error (MAE)
The average size of a model's misses when it guesses numbers. Lower is better.
Root Mean Squared Error (RMSE)
An error score for number guesses that punishes big misses more. Lower is better.