Abnormal Profiles

Abnormal Profiles is a standard configuration interface for adding, editing, and removing the profiles used by the Abnormal Thresholds and the Metric Abnormal Threshold Engine. The engine uses an aberrant detection algorithm to predict future deviation using the measures computed from past values and will alert based on them.

Refer to the Standard Configuration Interface guide for details on interacting with the grid and form.

This user interface calls REST methods from api/metric/abnormalProfiles.

To open this interface, from the main navigation menu, select Configuration, then Metrics, then Thresholds, and then Abnormal Profiles.

Form Fields

• Name - Name of abnormal profile

• Alpha - The degree to which the confidence bands smooth a weighted average (more weight being given to recent observations, less to older observations). Values are from >0 to 1. The closer to 1 the Alpha value becomes, the closer the predicted value will be to the weighted average of the last n observations.

• Beta - The degree to which the smoothing function considers the slope (amount of increase or decrease) of the weighted average of 2 adjacent data points. Values are from >0 to 1. The closer to 1 the Beta value gets, the more the Algorithm will consider the slope of the previous (n-1) data point to the current data point (n) when predicting the next data point (n+1).

• Gamma - The degree to which the smoothing function considers seasonal data when forecasting a data point. Values are >0 to 1. The closer to 1 Gamma gets, the more the Algorithm will weight seasonal patterns when forecasting expected values.

• Delta - Scaling factor, used to influence the width of the confidence bands generated by the Algorithm. Values are between 2 and 3.

• Epsilon - Values are >0 to 1.

• Window Length - Used with Violation Threshold by the Abnormal Thresholding Engine to determine whether to generate a threshold violation event. If Violation Threshold data points fall outside the confidence band within a Window Length number of points, then the data points are considered "abnormal" and an event will be generated. Values are between 0 and 28.

• Violation Threshold - Used with Window Length by the Abnormal Thresholding Engine to determine whether to generate a threshold violation event. If Violation Threshold data points fall outside the confidence band within a Window Length number of points, then the data points are considered "abnormal" and an event will be generated.

Best Practices

• Initially, it is best to use the default values and alter them once they are compared with actual data, as until there is a baseline determined for metrics, there is no definitive method of determining this information prior to metric base-lining.

• Delta can be calculated by (PercentWindowSize / 5.4%). For example:

  • 10% confidence band size / 5.4% = 1.85 as "Delta"

  • 65% confidence band size / 5.4% = 12.037 as "Delta"