Forecaster algorithms incorporate various features to improve Forecast accuracy. Feature contribution values help interpret forecast results. 

Explainability feature contribution values may be generated for underlying algorithm drivers. For example:

  • Related data line items
  • Historical data for the target value (actuals)
  • Seasonal and trend components

The explainability feature contribution values represent the relative importance of the features in percentage terms. The absolute value of the sum of all feature contribution values for a given item should add up to 100%.

Positive feature contribution values

For the Feature Contribution percentage, a value greater than zero means there's a positive relationship between explainability data features and the historical data. As the value of the related data line item increases, the value of the historical data increases as well. Examples:

  • A positive feature contribution percentage for a lagged value suggests that the time series (past values) correlate with future values.
  • A positive feature contribution percentage for a seasonal component means a positive impact on ‌the target value.

Negative feature contribution values

For the Feature Contribution percentage, a value less than zero means a negative relationship between the explainability feature and the historical data. As the value of the related data increases, the value of the historical data decreases. Examples:

  • A negative feature contribution percentage for a competitor's sales means that their increased sales are associated with a decrease in your sales.
  • A negative feature contribution percentage for a seasonal component means that the season has a negative impact on the target value.