The forecasting feature provides time series data modeling and forecasts based on data
presented in corresponding visualizations.
The forecasting feature is controlled by the
Forecast dialog box available
at the right corner of each eligible visualization. A visualization is eligible if both the
visualization type and specified data are supported for forecasting.
Note: Support for forecasting
is available for packages only and is not available in data modules containing a Planning Analytics
cube. However, you can use the forecasting feature with a Planning Analytics cube if you create an
enriched package from the cube. For more information, see
Creating an enriched package from a Planning Analytics cube.
When the Forecast dialog box is available and the feature is turned on,
appropriate time series models are estimated and computed forecasts are displayed in the
visualization. The Forecast dialog box contains user controls to assist in
generating suitable models, forecasts, and corresponding confidence bounds. All results are updated
when any of the controls are adjusted. While forecasts and confidence bounds are displayed or
available in the visualization, the corresponding time series model’s specification and data
processing summary are available in Forecasting statistical details found in the data tray below the
visualization.
The tooltip corresponding to a future value displays the corresponding time point, the forecast
value, as well as upper and lower confidence bounds for the forecast value. Forecasting values and
confidence bounds appear in a visualization.
Procedure
-
In a visualization that supports forecasting, click the Forecasting icon
.
-
Depending on the visualization, the following forecasting options are shown:
- Forecast periods
- The number of steps to forecast ahead.
- The default value is Auto, which is 20% of the length of the historical
data. Any missing values at the end of a particular series will also be forecast, but they will not
count towards the specified number of forecast periods.
- Ignored last periods
- Ignores a specified number of data points at the end of a time series when building the model
and computing the forecasts. Any missing values at the end of a non-ignored portion of a series will
also be forecast. Ignored last periods value must be specified as a non-negative integer, such as:
0, 1, 2, 3.
- The default value is 0. If there are no missing values, then all of the historical data is used
in model generation and the first forecast point is after the last historical data point. Up to 100
data points can be ignored.
-
Ignoring the last data period can be useful when the data is incomplete. For example, you might
be doing a forecast halfway through a month. Exclude this month from the forecast by setting
Ignored last periods to 1.
The following visualization shows a forecast that ignores September's results by setting
Ignored last periods to 1.
- Confidence level
- The certainty with which the true value is expected to be within the given range. You can see
corresponding confidence interval in a tooltip by hovering over any forecast value. The confidence
interval is displayed as upper and lower bounds.
- You can select 3 different confidence levels: 90%, 95%, and 99%. The default is 95% and the
lower and upper bound define the range at which you can be 95% confident that the true value lies
within that range.
- Seasonal period
- The seasonality with which to build the model. Seasonality is when the time series has a
predictable cyclic variation. For example during a holiday period each year.
The default value is
Auto. Auto automatically detects seasonality by
building multiple models with different seasonal periods and choosing the best one.
- You can specify seasonality by entering a non-negative integer, such as: 0, 1, 2, 3 as the
seasonal period.
- To specify a non-seasonal model, set the Seasonal period to 0 or 1. A
model with user specified seasonality is displayed only if the seasonal model is more accurate than
all of the non-seasonal models