Tip: | To preview these steps, work through Tutorial 1—Shampoo Sales. |
Follow these steps to set up a Predictor forecast and generate results:
Create and open a spreadsheet model with historical data as described in Creating Spreadsheets with Historical Data.
Select a data cell and start Predictor (see Starting Predictor and Running a Forecast).
Note:
You can select an entire data range or a single cell and let Predictor determine the range. If columns or rows of data are separated by blank columns or rows, you can use Ctrl+click to select one cell in each data series. For details, see Selecting Discontiguous Data.
Display the Input Data panel of the Predictor wizard.
If Welcome opens, click Next to display Input Data.
The appropriate data range is selected, including any row labels and column headers
Column Header and Label settings are correct
For details, click Help or see Selecting the Location and Arrangement of Historical Data.
In Data Attributes, indicate the time period for the data.
For example, if the data points represent monthly numbers, select months.
For Seasonality, select AutoDetect so Predictor will use statistical algorithms to determine whether the data is seasonal. Findings are displayed in a statement to the right of the list box. To fine-tune seasonality settings or use optional events and screening settings, see Selecting Data Attributes—Seasonality, Events, Screening.
Optional: If you are analyzing more than one data series with AutoDetect, click View Seasonality to chart the seasonality for each series.
For more information, see Viewing Historical Data by Seasonality.
Click Next to open the Methods panel, and select forecasting methods.
Depending on the Data Attributes Seasonality setting, select one or more of these:
Non-seasonal Methods—Work best on data that does not show a pattern that repeats regularly over a certain number of time periods, but can show a trend of decreasing or increasing over time
Seasonal Methods—Work best on data that shows a pattern that repeats regularly over a certain number of time periods and can also show a trend of decreasing or increasing over time
ARIMA—Useful in a variety of situations, particularly with many historical values and very few outlier values
Multiple Linear Regression—Useful when independent variables affect another variable of interest
Tip:
If Non-seasonal Methods and Seasonal Methods are available, select both.
If you have selected several series and one of them is controlled by the other, it is a dependent variable. In that case, select Multiple Linear Regression and see Using Multiple Linear Regression.
When settings are complete, click Next to review or change forecasting options.
Select an error measure and a forecasting technique.
The Glossary in this document and the Predictor sections of the Oracle Crystal Ball Reference and Examples Guide describe these settings. For basic forecasting, use the defaults: RMSE and standard forecasting.
When all Options settings are complete, click Run to run the forecast and produce results. For more information, see Starting Predictor and Running a Forecast.
The following topics describe how to customize Predictor settings to more closely reflect the historical data and provide more accurate forecast results: