Analyzing Advanced Prediction Results

Depending on how your administrator set up Advanced Predictions, you can review forms and dashboards to analyze Advanced Prediction results, and use Explain Prediction to review more details about the prediction results.

This topic shows some of the options that might be available to you, depending on your implementation.

To review Advanced Predictions:

1. Navigate to the Advanced Predictions results form or dashboard as directed by your administrator.
2. Depending on the implementation, you can use a dashboard to see an overview of the prediction, including historical data trends, and review the historical impact of various external and input drivers, such as industry volume, advertising and promotions, sales price, and GDP rates on historical sales volume.
   
   

3. Depending on the implementation, you can also review:

   • Prediction—Review prediction details, historical data, and input driver details.
   • Input Drivers—Review and edit the input driver details. You can review historical and predicted input driver values. If you edit the input drivers, an administrator can re-run the prediction to get updated prediction results.
   • Forecast Accuracy—Review and ensure the accuracy of the prediction.
   • Predict by Algorithms—Explore prediction results using different algorithms.

4. To get more information about the prediction, in a form with predicted values, right-click a cell with predicted data and then select Explain Prediction.

   Note:

   Explainability is available only in Forms 2.0 on the web. It is not available in dashboards or in Oracle Smart View for Office.
5. Review the information on the Explainability screen:
   
   

   • The historical data is shown as an orange series to the left of the vertical separator line.

   • The base case (most likely case) for predicted data is in the purple series to the right of the vertical separator line.

   • The predicted data series is bounded by dotted lines that show the confidence intervals (the upper and lower prediction intervals)—the range between the Best Case predicted scenario and the Worst Case predicted scenario.

   • The Model estimates line, also known as the fitted value line, shows the ML model’s estimates for historical data based on its learning of underlying logic and trends. Comparing the fitted values with historical actuals data indicates how well the prediction model was able to capture the variations in the provided data. The prediction is made using the selected algorithm for future values using the fitted values of historical data.
6. You might have the option to refine input drivers. Then, ask your administrator to run the Advanced Prediction job again. Review the impact of your changes.
7. If your administrator implemented an option to review Forecast Accuracy, you can compare the accuracy of the prediction results using the Forecast Value Add (FVA) metric. In this example, you can compare the Advanced Prediction results (Multivariate Prediction), Univariate Prediction (results from Predictive Planning) and Forecast to backtesting that is performed to measure the prediction accuracy based on the historical data.

   To calculate FVA, the accuracy of the adjusted forecast is compared to the accuracy of a baseline. If the adjusted forecast reduces errors compared to the baseline, then it has a positive FVA; if it increases errors, the FVA is negative. This metric helps you focus on the steps that improve accuracy and eliminate non-value-adding activities in the forecasting process.