The primary challenge in demand forecasting is to plan a forecasting strategy that minimizes forecasta numerical prediction of a future value for a specified time period for each unique combination of BY variable values error. By using the available demand history, you can get in-depth information about the historical demand patterns for a time seriesan aggregation of transactional data into specified time intervals and sorted according to unique combinations of the default attributes (BY variables). SAS Visual Forecasting enables you to classify historical demand patterns, group the time series based on certain criteria, and then apply the most suitable modeling techniques based on the classification for each time series.
For example, a grocery store might sell both regular candy and holiday candy, which have different demand patterns. You might segment the candy based on the time span when it is sold. Candy that is sold throughout the year has a long time span. Seasonal candy (for example, Valentine's Day chocolates) has a short time span. To generate a modeling node that produces accurate demand forecasts, products must be segmented appropriately, based on their demand patterns. Demand classification provides an automatic and data-driven way to handle segmentation and modeling.
When a forecasting project is created, each time series is classified based on characteristics such as high volume, low volume, seasonalitya regular change in time series data values that occurs at the same point in each time cycle., time span, and such. These classifications are shown on the Data tab as Demand Classification attributes. These attributes can be used to filtera set of specified criteria that are applied to data in order to identify the subset of data for a subsequent operation, such as continued processing. the time series when using the Time Series Viewer and Forecast Viewer. You can also create overrides based on the demand classification filters. In addition, you can use the Demand Classification pipeline template to segment the time series and use modeling nodes targeted for their specific classifications.