Forecast creation

For testing purposes, a forecast can be created using the forecasting screen. However, in a production setup, the forecasts are typically created using one or more batch jobs that are scheduled in the job scheduler. These jobs are usually set up to run on a weekly basis, creating forecasts for the entire store population in one task.

Forecast from historical results

When you generate a forecast from historical results, the system generates an initial forecast (the system forecast) based on historical trends. You can then manually adjust the forecast to account for other events as necessary. This adjusted forecast becomes the manager adjusted forecast.

The location’s forecast method determines which historical trends are used to create the forecast. These options are available:

  • Linear Regression - Data from the specified historical period is used.
  • Trend of Historic Averages - Data from the previous eight weeks of actual data is compared with the same weeks last year to calculate a day-by-day and year-over-year trend. The trend is then applied to the actual results from the forecast period in the previous year. If there are six or more days of data available within the eight weeks, the highest week and the lowest week are dropped during the calculation.
  • Multiple Regression - A historical relationship trend is applied to a known imported forecast value to create the forecast.
  • Machine Learning - The machine learning forecasting model uses forecasting models created during the training process to predict results for the forecast period. These forecasting models are generated based on an analysis of historical actual results and other data that is relevant to the location that is being forecasted.