Creating a Forecast - OLAP

The OLAP DML supports simple linear regressions, several non-linear regression methods, single exponential smoothing, double exponential smoothing, and the Holt-Winters method. If you are unsure of which method to use, you can have the OLAP engine decide the best fit for your data based on past performance. Most forecasts are calculated at the base level. You then aggregate the base-level forecast data to generate forecast aggregates. Typically, you do not generate forecast aggregates from the aggregates of actual data. The examples in this chapter assume that you wish to generate forecast aggregates in this way.
However, at times you may want to generate forecasts at the aggregate level and then allocate the data to lower levels. This method of forecasting is also supported.

Steps for Creating a Forecast

These are the steps for creating a forecast. Each one is discussed in more detail in the sections that follow.

  1. Verify that the time periods for the forecast have been created in your time dimension. Add them if necessary.
  2. Define the variables that will be used to store the results.
  3. Write a program that generates the forecast.
  4. Compile and run the program.
  5. Check the results.
  6. Add the results measure to a cube. Optionally, first create a new cube for forecasting results.
  7. Create a new aggregation plan or modify an existing one to include the measure containing the forecast results. Deploy the aggregation plan.
  8. Enable the analytic workspace for your applications.

Creating the Forecast Time Periods

The future time periods that you want to forecast must be defined as members of the time dimension in your analytic workspace. If they do not exist there already, you must:

  1. Add the new members and their attributes to the Time dimension table in the source schema.
  2. Use the Refresh wizard in Analytic Workspace Manager to add the new members to the dimension in the analytic workspace.

You should use whatever mechanism guarantees that these Time dimension members will be identical when you load actual data.

Defining Variables for the Results

A forecast requires a minimum of one variable for the results, and up to three variables if you want seasonal and smoothed seasonal forecasts. These variables typically have the same dimensions and data type as the variable used to generate the forecast.
Take these steps to define the variables for a forecast:

  1. Define the results variable as a standard form measure.
  2. For a seasonal forecast, define a second variable for the seasonal factors. Do not assign standard form properties to this variable. Instead, do the following:
    • In the Object View, expand the folder for your analytic workspace.
    • Right-click Variables and choose Create Variable from the menu.
    • Define the variable with a DECIMAL data type.
    • On the Dimensions page, list the dimensions in the appropriate order for variables in your cube, typically Time first, then a composite dimension.
  3. For a smoothed seasonal forecast, define a third variable for the smoothing factors. Copy the seasonal factors variable by right-clicking the variable and choosing Create Like.

All rights reserved © 2018 Wisdom IT Services India Pvt. Ltd DMCA.com Protection Status

OLAP Topics