Concept: Seasonality for Predictive Forecaster
Definition of Seasonality
Seasonality is a term used to describe fluctuations in data that follow a cyclical pattern. The algorithm uses the historical data and parameters that you provide to determine seasonal peaks. It then assumes that the same seasonal pattern will repeat in the future. Some of the algorithms available in Predictive Forecaster detect seasonality automatically, and others enable you to define the number of periods that constitute a season.
Custom Seasonality with Predictive Forecaster
Prophet, Holt-Winters, and Orbit DLT enable you to customize the seasonality. You can also customize seasonality for AutoFit, which applies when
AutoFit
uses 1 of the algorithms that support the customization.The Kalman algorithm is unique in that you can indicate whether or not you want the algorithm to apply seasonality to the forecast without having to define it yourself.
You can allow our ML system to automatically detect seasonality in your reference data, or you can configure seasonality manually.
Adaptive Planning provides these options for seasonality:
Seasonality Option | Descriptions |
|---|---|
Yearly | Yearly seasonality is the default for seasonality. Use when you have:
|
Biannual | Use when you have:
|
Quarterly | Use when you have:
|
Weekly | Available for Prophet and AutoFit only.
Use the Add weekly seasonality check box in addition to Yearly , Biannual , or Quarterly seasonality. Weekly seasonality is best for data:
|
Multiplicative | Available for Prophet , Holt-Winters , and AutoFit .
Use the Trend varies greatly with seasonality check box when seasonal spikes and drops vary in proportion to the trend. |
None | Use when:
|