MAE (Mean Absolute Error)

Mean Absolute Error (MAE) is a measure of the average absolute magnitude of the errors in a set of predictions, without considering the direction of the errors. It is calculated by taking the absolute value of the difference between each prediction and the actual value, and then averaging these absolute differences. Unlike MSE or RMSE, errors are not squared here. This means MAE does not penalize outliers as heavily as squared error metrics. MAE has the same unit as the target variable, making it easily interpretable. A lower MAE indicates a better model.

For marketing leaders, MAE is relevant when a robust metric for prediction accuracy is needed, especially in the presence of outliers. When forecasting marketing costs, advertising spend, or customer response rates, where individual extreme values can occur, MAE provides a more stable assessment of average prediction accuracy. It helps understand the typical deviation without being dominated by individual extreme errors.

A company uses a model to predict the number of daily website visits. Occasionally, extreme peaks occur on certain days. An MAE of 500 visits would mean that the model's predictions deviate by an average of 500 visits from the actual values, without rare, extremely high deviations distorting the overall picture.

MAE is one of the oldest statistical metrics, already in use in the 18th century.

• Scikit-learn: MAEDocumentation