The mean absolute error (MAE) of predictions 
for observed values 
is the arithmetic mean
of the absolute errors 
, where 
is the signed error. Thus
 |
It has the same units as the observations and, unlike the mean square error, weights errors linearly rather than quadratically.
