Entering content frame

Process documentation Quality Improvement of an MLR Model Locate the document in its SAP Library structure

Purpose

In this process, you improve the accuracy of forecasts created with multiple linear regression by monitoring predefined tolerance thresholds for the MLR measures of fit using alerts, and by adjusting the forecast model where any of these thresholds is exceeded.

There are two situations in which you use alerts to monitor the MLR measures of fit:

Prerequisites

You have installed the Alert Monitor.

Process Flow: Creation of an MLR Model on the Demand Planning Desktop

  1. Set the upper and lower limits of the MLR measures through the diagnostic group in the MLR profile. For more information, see Measures of Fit.
  2. In the Alert Monitor, create one or more forecast alert profiles for Demand Planning.
  3. Alert profiles are used to display specific alerts to specific users. They do not affect whether or not alerts are created.

  4. In interactive demand planning, assign the alert profile you created in step 2 by choosing Edit → Assign alert profile. Do not select Forecast deletes old alerts automatically.
  5. Run the forecast in the multiple linear regression view of interactive demand planning.
  6. The MLR measures of fit are calculated automatically. Durbin-h, the t-test and mean elasticity are calculated for each variable. The other measures are calculated for the model as a whole. You see them in the multiple linear regression view by choosing Measures of fit.

  7. From the workspace toolbar, choose Alerts on/off This graphic is explained in the accompanying text.
  8. The alerts appear at the bottom of the screen.

  9. Check the alerts to see which measures of fit have exceeded their predefined upper and/or lower limits.
  10. Make any necessary adjustments to the MLR model, for example, by deleting an explanatory variable or entering a new one.
  11. Run the forecast again.
  12. Check the alerts to see whether the adjustments you made in step 7 have corrected the problem.
  13. Message alerts that you viewed in step 6 are still visible, so you can compare them with the new ones. This is the case if you did not select Forecast deletes old alerts automatically in step 3.

  14. Repeat steps 7 through 9 until no more new alerts appear in the Alert Monitor.

Process Flow: Running MLR with Mass Processing using a Preconfigured MLR Model

  1. Set the upper and lower limits of the MLR measures through the diagnostic group in the MLR profile.
  2. In the Alert Monitor, create one or more forecast alert profiles for Demand Planning.
  3. Alert profiles are used to display specific alerts to specific users. They do not affect whether or not alerts are created.

    In the mass processing activity, specify that alerts are to be created.

  4. Run the forecast with mass processing.
  5. The MLR measures of fit are calculated automatically. Durbin-h, the t-test and mean elasticity are calculated for each variable. The other measures are calculated for the model as a whole.

  6. In the Alert Monitor, see if any measures of fit have exceeded their predefined upper and/or lower limits.

If alerts have been issued, this is an indication that you need to make further adjustments to the MLR model - at least, for some products. For these products, resume the quality improvement procedure described above.

Leaving content frame