MAE

Also known as · Mean Absolute Error

Mean absolute error, MAE=1n∑i∣yi−y^i∣\text{MAE} = \frac{1}{n} \sum_i |y_i - \hat{y}_i|: the average size of the prediction errors, in the same units as the target.

When to use

Choose it over RMSE when Outliers exist or interpretability is key, since it is less sensitive to large errors; RMSE penalises big errors more heavily. Used in PS 02 to compare regression models.

See Lec 4.

Appears in

No references yet.