Mean Absolute Error in BigQuery ML

What Mean Absolute Error (MAE) is:

Mean Absolute Error (MAE) is a fundamental evaluation metric used in regression tasks to measure the average magnitude of errors between predicted values and actual target labels. It is calculated by taking the arithmetic average of the absolute differences across all observed and predicted data points, treating over-predictions and under-predictions equally.

How it meets the requirement:

• Measures average absolute difference (✓): MAE quantifies the typical prediction error in the exact same physical units as the original target variable (e.g., dollars, units, or temperatures).
• Lower value indicates better performance (✓): MAE ranges from zero to infinity ($[0, \infty)$). A score of 0 represents a theoretically perfect predictor, and lower non-zero values reflect smaller errors and higher model quality.
• Supported in BigQuery ML regression evaluations (✓): When executing ML.EVALUATE on a regression model, BigQuery automatically returns mean_absolute_error alongside other standard regression metrics.

Key advantages and capabilities:

• Direct interpretability: Because errors are not squared, MAE provides an intuitive, linear assessment of average model error.
• Robustness to outliers: Unlike squared error metrics, MAE does not disproportionately penalize occasional large outlier errors.
• Scale representation: Outputs error values on the original scale of the continuous target column.
• Benchmark comparisons: Allows clear comparison between different candidate regression models evaluated against the same baseline dataset.

Why this is optimal:

MAE directly satisfies the criteria of calculating the average absolute error magnitude where smaller scores represent superior model accuracy, making it the correct metric for regression model evaluation in this scenario.