Weighted Mean Absolute Percentage Error

Weighted Mean Absolute Percentage Error (WMAPE) refines MAPE by weighting errors based on actual values, addressing imbalances in datasets where some observations are more critical. This metric is valuable for skewed data, ensuring errors on important points influence the score more. Lower WMAPE signifies better weighted accuracy, highlighting proportional errors with emphasis on higher-magnitude instances.

The formula for WMAPE is:

$$\text{WMAPE} = \frac{\sum_{i=1}^n |y_i - \hat{y}_i|}{\sum_{i=1}^n |y_i|}$$

Often expressed as a percentage by multiplying by 100, where$y_i$ are actuals and$\hat{y}_i$ predictions, normalizing by the sum of absolutes for robustness.

WMAPE enhances forecasting reliability, particularly where high-stakes periods matter. It mitigates MAPE's issues with zeros or low values by aggregation.

WMAPE's focus on magnitude makes it ideal for volume-driven variables, complementing other metrics for holistic evaluation. However, it assumes actuals as weights—customize if alternative weighting is needed.