Create an alarm rule

Alarm rules can be easily created in Anomaly by following the procedures below:

Select a data source

Follow the procedure below to create an alarm rule:

  1. Click Create Alarm Rule on the upper right of Anomaly.

    create rule
  2. Select the data source you want to monitor.

    select datasource

Select a measure(s) to monitor

Once a data source is selected, the following page is displayed with the Data panel opened on the left. Use this panel to select a measure(s) to monitor as follows:

  1. Under the Measure area, choose a column to apply an alarm to. The selected column is automatically placed on the Aggregate shelf.

    select measure
  2. If required, a custom column can be created by applying an expression to an existing column. On the upper right of the Measure area, click icon_custom_column and set up a custom column in the dialog box.

    make custem measure rule
  3. Click the aggregation type menu of each column on the Aggregate shelf and select a desired type.

    select aggregation type
  4. If required, aggregate data can be split based on a dimension. Under Dimension, hover the cursor over the column you want to use as a splitter, and click icon_split.

    splite measure by dimension
  5. If required, aggregate data can be filtered based on a dimension. Under Dimension, hover the cursor over the column you want to use as a filter, and click icon_ano_filter. Next, select the elements you want to monitor.

    select filter value

Specify a training interval

Once a measure(s) is selected, open the Training interval panel and choose a range of data to be used for prediction model training.

  1. Choose a training interval for the model under Granularity.

    select granularity
  2. Set a date range for data to be used in training.

    select specific time
  3. Click Next to continue.

Select a model

Now, proceed to the Model panel and choose a prediction model. Anomaly trains each model using the given training data set and calculates the results. Below are two methods of choosing an appropriate prediction model.

  • The model with the highest accuracy (out of 100 points) is automatically displayed on the right, and marked as Recommended.

    auto training model
  • Hovering the cursor over each item provides detailed information, and you can select the desired model.

    model details

Set alarm rule conditions

After selecting a prediction model, set up conditions for alarm triggering in the Condition panel.

  1. Click icon_set on the right of Subscribers to open the following dialog box, and set up subscribers and method of notification.

    set alarm noti detail
  2. Read the description below, and set up the alarm time and interval.

    set alarm interval
    • Alarm Start: The alarm feature is turned on at this time. Alarms can be triggered beginning from the start time.

    • Alarm Interval: When a condition is met, alarms are set off at the intervals set by this field.

  3. Read the description below, and set up trigger conditions based on abnormal scores of data being monitored. By default, one condition is given, and more conditions can be added by clicking + Add Condition.

    set alarm condition
    • Severity: Set up the alarm severity corresponding to the given condition.

    • Threshold: The monitored data is considered abnormal when this setting is exceeded.

    • Frequency: Set how many times the abnormal score should exceed the threshold within the specified period to set off an alarm. For example, “3 within 5 minute” means that an alarm is set off if the abnormal score exceeds the threshold 3 times within 5 minutes.

  4. Click Next to continue.

Complete the alarm rule

Once the alarm rule is set up, proceed as follows:

  1. Enter the name and description of the alarm rule, and click Done.

    write alarm rule name
  2. The newly created alarm rule is displayed on the top of the alarm rule list. The current status remains as Prepared up to the first alarm.

    alarm rule list