3556469 - Setting Threshold for Predictive Categories in SAP Analytics Cloud

• In SAP Analytics Cloud, a question arises regarding the ability to set a custom threshold for predictive categories

Currently no direct way to set a decision threshold explicitly. A default threshold is always used for decision, which is determined such that the percentage of positive predictions is identical to the percentage of positive cases observed in the training dataset.

1. To workaround this limitation, use the prediction probability (representing the probability for an observation to belong to the positive target class) as output instead of the decision.
   How is a Decision Made For a Classification Result?
2. In a story, create a new calculated measure representing the decision using a formula similar to IF(decision_probability >= THRESHOLD, 1, 0) where THRESHOLD is replaced by the threshold of choice.
3. Using a calculated measure to calculate the decision allows experiment with threshold values without having to apply the predictive model for each single experiment.

SAP Analytics Cloud, Predictive Categories, Threshold, Prediction Model, Positive Contributions, Negative Contributions, Training Data, Decision Threshold, Prediction Probability, Calculated Measure. , KBA , LOD-ANA-PR-SP , SAC Predictive Smart Predict , Problem