MVLR is used for feature-based forecasts. You can use MVLR when:
- There are multiple features, such as promotions or inventory levels
- Features may be related to the SKU (promotions or inventory levels)
- Features may be unrelated to the SKU (oil prices or gross domestic product)
The MVLR algorithm uses feature data to generate a prediction. Features are important when they influence prediction results.
MVLR feature input sources
- Historical data
- Related data
- Synthetic data:
- These features are created by PlanIQ automatically. They can be based on either historical or related data. Examples are:
- Exponential trends
- Linear trends
- Seasonality
- Lagged values
(for example, a gross margin prediction, based on a fuel price change three weeks prior)
- These features are created by PlanIQ automatically. They can be based on either historical or related data. Examples are:
At the "create a related data module" stage, for the MVLR algorithm only, you can include up to 28 related time series (RTS) columns.