Concept: Anomaly Detection in Sheets

Anomaly detection starts with your actuals data. Based on the actuals data, the algorithm creates a model of predicted values that it stores in a designated machine learning forecast version. The algorithm then creates a range of accepted data that compares the machine learning forecast against any other plan version. Sheets highlight the data that fall outside the prediction range. You can also build outlier reports.

Sheets highlight anomalies for rollups differently depending on the rollup type:

Rollup Type	Behavior Towards Anomalies
Accounts	The algorithm doesn't detect out of range data in the account rollup. However, the sheet highlights the account rollup when there’s out of range data in the contributing accounts. When an account rollup cell has a purple border, expand the account to find the highlighted anomalies.
Splits	The algorithm detects and highlights out of range data in split rollups. Expand and change the data of individual splits to bring the total within range. The sheet doesn't highlight the contributing splits that are out of range.
Time, level, and dimension rollups	The algorithm doesn't detect out of range data in these rollups. The sheet also doesn't highlight the rollups when: The total is out of range. The contributing data are out of range. Keep these rollups expanded to find highlighted anomalies.