Read anomaly scores

An anomaly score is provided with each news story generated by Guided Analytics. A score of 0 indicates that the model sees normal conditions based on the provided training data, while a score of 100 indicates that the data being received matches the most anomalous data provided in the training sample. Stories are generated and published to the news feed once the anomaly score exceeds 85.

Scores exceeding 100 indicate a major disturbance that exceeds even the most anomalous data from the training sample and should be promptly investigated.

In addition to using anomaly scores to gauge current operating conditions in your solution, the score can be used as a way to calibrate the sensitivity of the model. For example, if you desire a model that is highly sensitive to relatively small deviations from normal operating conditions, the provided data for model training should contain only data indicating normal behavior. If you desire a model that is less sensitive, ensure that the training data is varied and includes both normal operating data as well as minor disturbances.

While you should always avoid including data that presents major disturbances, a varied data set can better define the model and avoid unnecessary alarms.