Skip to main content
Documentation
Documentation
Adaptive Planning
Last Updated: 2023-06-23
FAQ: Anomaly Detection

FAQ: Anomaly Detection

What do I need to do to get started?
First create a machine learning prediction version. Then generate predictions. See Create Prediction Versions for Anomaly Detection.
What does the prediction version do?
The prediction version holds predicted data calculated by the machine learning algorithm.
Can I edit or delete the prediction version?
You can edit some of the settings. Adhere to required version settings. The algorithm can fail if you don't.
You can also delete the version. When you delete the version, anomaly detection and outlier reporting stop working.
You can't edit the data in sheets for the prediction version because the version is read-only.
What kind of sheets display anomalies?
Only standard sheets that don't have custom dimensions.
What kind of accounts display anomalies?
  • Only periodic general ledger and custom accounts that aren't linked to other accounts.
  • The
    Data Entry Sheet Type
     account setting must be
    Standard
    .
  • Collapsed account rollups are marked anomalies if at least 1 child account has unexpected data. You can expand the account to see where the anomaly is.
What kind of versions display anomalies?
Only plan versions display anomalies.
Can I display anomalies in Excel Interface for Planning sheets or OfficeConnect reports?
At this time, anomalies detection works only in Adaptive Planning sheets and web reports.
How does Adaptive Planning detect anomalies?
We use a predictive algorithm that studies the patterns of your actuals version. By default, we define upper and lower bounds for each prediction as follows:
  • UpperBound = PredictionValue + 0.4 * | PredictionValue |
  • LowerBound = PredictionValue - 0.4 * | PredictionValue |
When values are greater than the upper bound or less than the lower bound, they display as anomalies. You can change the thresholds with the AIML-THRESHOLD sheet. See Define Anomaly Thresholds.
When do predictions start?
Use the
Start of Plan
 setting in your prediction version to define when predictions start. Then, the algorithm predicts values for the next 24 time periods.
What if my sheet isn't showing anomalies?
An account needs at least 24 data points in the actuals version before the algorithms start detecting anomalies in that account.
Also, make sure you're:
  • Looking at general ledger or custom accounts. Cube, modeled, metric, and assumption accounts don’t display anomalies.
  • Looking at periodic accounts. Cumulative and link accounts don't display anomalies.
  • Displaying the smallest time period. Example: Months instead of quarters. The algorithm doesn’t detect anomalies for time rollups.
  • Looking at a level with no child levels. The algorithm doesn’t detect for level rollups.
  • Looking at a standard sheet with the
    Show Anomaly
     button. The button doesn’t display when your standard sheet has custom dimensions.
  • Looking at a plan version. Actuals versions don't display anomalies.
  • Not looking at the prediction version, which holds the predictions, which means they’re never anomalies.
  • Not looking at splits.
What can I do if the anomalies detected aren't anomalies?
  • Check your actuals version to ensure that data is accurate. The algorithm counts blank cells as zeros, which can affect predictions.
  • Consider defining your own thresholds with AIML-THRESHOLD sheet.
  • To remove the anomaly border from a cell, right-click on the cell and select
    Unmark anomaly
    .
What does outlier reporting do?
This matrix report uses the predicted values in the prediction version. Using standard report functions, you can create outliers based on your custom-defined upper and lower bounds. These outliers don’t match the anomalies you find in sheets because the report formatting defines them, not the algorithm. See Step: Build Outlier Reports.
Do the upper and lower bounds in my outlier report update detected anomalies in sheets?
No. The anomalies in sheets use the upper and lower bounds defined by the machine learning algorithm. The outlier report offers a way for you to analyze the differences between your plans and the prediction of the algorithm. It can be useful to you, but doesn’t change the definition of anomalies for the algorithm.