Notes by Lex Toumbourou

Mean Absolute Error (MAE) is a metric for assessing Regression predictions. Simply take the average of the absolute error between all labels and predictions in the test set:

$\frac{1}{N}\sum\limits_{j=1}^{n} |y_i - \hat{y}_i|$

Step-by-step:

1. Calculate error vector as labels - predictions
2. Take the absolute values of the errors
3. Take the mean of all values.

It's also known as L1 Loss or L1 Penalty because it takes the l1-norm of the error vector

An alternative to Root Mean-Squared Error.

References

• What Are L1 and L2 Loss Functions

Backlinks

• ML Regression
• Root Mean-Squared Error