Forecast methods

Machine Learning

The machine learning forecasting method creates a forecast based on a machine learning analysis of historic volume data and additional data points provided by the customer or from third-party services. The machine learning method is suitable for customers that have historical sales or other volume data spanning multiple years. This model requires additional configuration steps to ensure that models can be built for each location being forecast.

See Forecasting with machine learning.

Trend of Historic Averages (ToHA)

Trend of Historic Averages (ToHA) forecasts sales or transaction volumes by comparing the trend of this year's actual data against the trend of last year's actual data. For each day, a forecast is calculated by multiplying the actual data for the same day in the previous year by the ratio of the change between this year's performance and last year's performance. This forecast method accounts for seasonality by retrieving data from the current year and previous year during the same calendar period as the date being forecast. This data is taken from days that match the same day of the week as the date being forecast. This forecasting method requires a minimum of 13 months of historical data.

For more details, see Trend of Historic Averages algorithm.

Linear regression

Linear regression is best used when seasonality is not a factor or when there is insufficient historical data to use Trend of Historic Averages. In this forecasting method sales information from a historical period is used to forecast a future period. The historical period is selected during the forecast creation process. A linear regression algorithm then uses that historical information to forecast the upcoming week.

Inherited

This type of forecasting is rarely used. The forecast is inherited from another location. In this forecasting method you select a location from which to inherit a forecast. You can choose whether the inheritance type is fixed or dynamic. The fixed type is a static inheritance percentage. The dynamic type calculates the inheritance percentage based on the location’s historical data.

External Workload Feed

Schedules created for locations using the External Workload Feed forecast method have their interval requirements determined by the workload data recorded in the SO_PLAN_WORKLOAD database table.

The workload data is imported by an external system for each department daily, and distributed evenly across the department’s hours of operation. Forecast generation is not required because the forecast data is already provided.

The Historical data is contained within sub-locations for this location check box on the Locations Properties page must not be selected to generate data for a department. If you select the check box, you will generate data for the driver instead.

You must specify the same Forecast Interval Type for all locations and its sub-locations (that is, the department and its drivers).

For the External Workload Feed to function properly, you must provide at least 1 record for each Forecast Interval for a 24 hour period.

For example, if you specify a Forecast Interval of 15 minutes, you must provide at least 1 record for each 15 minute interval. Therefore, for a 24 hour period, you must provide at least 96 records in total (4 records per hour x 24 hours = 96). Similarly, if you specify an interval of 30 minutes, you must provide at least 1 record for each 30 minute interval, and therefore, at least 48 records in total (2 records per hour X 24 hours = 48).

For more information on the interface data required for this type of forecast, see the Infor Workforce Management Interfaces Administration Guide.

Imported

This forecast type allows for importing forecast values into the staging table SO_FCAST_EXTERNAL. This forecast type is required in conjunction with multiple regression, but the type may be used alone, or along with other forecast types. These values will then be used during forecast creation. For creating a generic table import to populate values into the staging table, see the Infor Workforce Management Interfaces Administration Guide.

Multiple regression

The multiple regression forecast method uses linear regression with multiple independent drivers to create a forecast for a single dependent driver. This method uses historical data to establish a relationship between the independent and dependent drivers. That relationship is applied to a forecast value that is imported for each dependent driver to create the forecast for the independent driver.

The calculations use the Ordinary Least Squares method of linear regression estimation. This method is the same as the Excel function “TREND”. To model in Excel, you can use this formula:

=TREND([dependent driver historical data],[dependent drivers historical data],[imported forecast for dependent drivers])

In the formula, the data should only include one day of the week at a time, and the historical data ranges should be for the same time period.

The precision differs between Excel and the Java model, so some differences may appear. This formula is a very close approximation of what is happening within the system.