Product documentation Docs
    • English
    • Deutsch
    • 日本語
    • 한국어
    • Français
  • More Sites
    • Now Community
    • Developer Site
    • Knowledge Base
    • Product Information
    • ServiceNow.com
    • Training
    • Customer Success Center
    • ServiceNow Support Videos
  • Log in

Product documentation

  • Home
How search works:
  • Punctuation and capital letters are ignored
  • Special characters like underscores (_) are removed
  • Known synonyms are applied
  • The most relevant topics (based on weighting and matching to search terms) are listed first in search results
Topics are ranked in search results by how closely they match your search terms
  • A match on the entire phrase you typed
  • A match on part of the phrase you typed
  • A match on ALL of the terms in the phrase you typed
  • A match on ANY of the terms in the phrase you typed

Note: Matches in titles are always highly ranked.

  • Release version
    Table of Contents
    • Now Platform capabilities
Table of Contents
Choose your release version
    Home Orlando Now Platform Capabilities Now Platform capabilities MetricBase Detect anomalies in MetricBase data using predictive models

    Detect anomalies in MetricBase data using predictive models

    • Save as PDF Selected topic Topic & subtopics All topics in contents
    • Unsubscribe Log in to subscribe to topics and get notified when content changes.
    • Share this page

    Detect anomalies in MetricBase data using predictive models

    Use statistical models to determine significant anomalies in real-time using MetricBase triggers. You will need to train a model using representative data that has already been stored in MetricBase.

    Before you begin

    Role required: admin

    About this task

    MetricBase creates a model by training a representative sample of your time series data to determine the model parameters. The training process determines the model parameters that best fit your data, to distinguish normal data from anomalous data. MetricBase supports the following model types:
    • Probabilistic Exponentially Weighted Moving Average (PEWMA), a moving average algorithm that uses a probability factor to determine how it reacts to change in data
    • Autoregressive Integrated Moving Average (ARIMA), a moving average algorithm that factors in previous errors and values
    • Seasonal Trend decomposition using Loess (STL), a seasonal algorithm for decomposing time series data into seasonal and trend components
    • Holt-Winters (HW), a seasonal algorithm that decomposes the trend and seasonal components to determine the level
    Note: MetricBase selects the most appropriate model type when you select Find Best Fit Model from the model class list.
    After you have a model trained from your data, you can trigger flows when new data is significantly different than the trained data.

    Procedure

    1. Navigate to MetricBase > MetricBase Models and click New.
    2. On the form, fill in the fields.
      Table 1. Model New Record form
      Field Description
      Model name Name of the model. The name can be any combination of alphanumeric characters. This model name is not the same as the model class. In general, the name relates to the value in Group by.
      Table name Name of the table that contains the training data.
      Metric Name of the metric that you use to train the model. The metric must belong to the table.
      Created Date that you trained the model.
      Filter Filters that you use to exclude some of the data in the dataset.
      Note: When choosing data to train your model, try to select data that demonstrates an expected behavior to reduce anomalies in the training set.
      Group by You can use Group by as a discriminator field for your model data. For example, if you want to create a data model over a group of production servers whose performance differs by role (such as database or application server roles) then you would choose role as the Group by field. The training process creates one model per role in the group of records selected by the filter. You do not have to manually create a model for each role.
      Model Class The algorithm to use when training data. Select a moving average algorithm (PEWMA, ARIMA), a seasonal algorithm (STL, HW), or choose Find Best Fit Model. The default is Find Best Fit Model, which tries each algorithm and selects the one which appears to have the best fit over the training set.
      Training Dataset Start Date MetricBase time series data for the metric starting with this date.
      Training Dataset End Date MetricBase time series data for the metric ending with this date.
      Valid until Date that serves as a reminder to consider retraining the model. If the model is performing well, there's no need to retrain it. The model can continue working past this date.
      Active Option to use the trained model. Once the model is active it becomes available for use as a Flow Designer trigger.
    3. Click Submit and Train.
      MetricBase trains the model. When complete, the model appears on the MetricBase Model Instances tab.
    4. Click the model name.
      The modeling data appears as does the model string with the parameters optimized by the training.
      Trained model data
    5. (Optional) Click the model name and then click Set Model to change the model parameters.
      You can edit the model parameters when you want to override the settings for training your model. The graph does not update, you are saving the revised model string.
      Model parameters

    What to do next

    You can create a Flow Designer trigger for this model. For more information, see Create a model trigger.

    Tags:

    Feedback
    On this page

    Previous topic

    Next topic

    • Contact Us
    • Careers
    • Terms of Use
    • Privacy Statement
    • Sitemap
    • © ServiceNow. All rights reserved.

    Release version
    Choose your release version

      Detect anomalies in MetricBase data using predictive models

      • Save as PDF Selected topic Topic & subtopics All topics in contents
      • Unsubscribe Log in to subscribe to topics and get notified when content changes.
      • Share this page

      Detect anomalies in MetricBase data using predictive models

      Use statistical models to determine significant anomalies in real-time using MetricBase triggers. You will need to train a model using representative data that has already been stored in MetricBase.

      Before you begin

      Role required: admin

      About this task

      MetricBase creates a model by training a representative sample of your time series data to determine the model parameters. The training process determines the model parameters that best fit your data, to distinguish normal data from anomalous data. MetricBase supports the following model types:
      • Probabilistic Exponentially Weighted Moving Average (PEWMA), a moving average algorithm that uses a probability factor to determine how it reacts to change in data
      • Autoregressive Integrated Moving Average (ARIMA), a moving average algorithm that factors in previous errors and values
      • Seasonal Trend decomposition using Loess (STL), a seasonal algorithm for decomposing time series data into seasonal and trend components
      • Holt-Winters (HW), a seasonal algorithm that decomposes the trend and seasonal components to determine the level
      Note: MetricBase selects the most appropriate model type when you select Find Best Fit Model from the model class list.
      After you have a model trained from your data, you can trigger flows when new data is significantly different than the trained data.

      Procedure

      1. Navigate to MetricBase > MetricBase Models and click New.
      2. On the form, fill in the fields.
        Table 1. Model New Record form
        Field Description
        Model name Name of the model. The name can be any combination of alphanumeric characters. This model name is not the same as the model class. In general, the name relates to the value in Group by.
        Table name Name of the table that contains the training data.
        Metric Name of the metric that you use to train the model. The metric must belong to the table.
        Created Date that you trained the model.
        Filter Filters that you use to exclude some of the data in the dataset.
        Note: When choosing data to train your model, try to select data that demonstrates an expected behavior to reduce anomalies in the training set.
        Group by You can use Group by as a discriminator field for your model data. For example, if you want to create a data model over a group of production servers whose performance differs by role (such as database or application server roles) then you would choose role as the Group by field. The training process creates one model per role in the group of records selected by the filter. You do not have to manually create a model for each role.
        Model Class The algorithm to use when training data. Select a moving average algorithm (PEWMA, ARIMA), a seasonal algorithm (STL, HW), or choose Find Best Fit Model. The default is Find Best Fit Model, which tries each algorithm and selects the one which appears to have the best fit over the training set.
        Training Dataset Start Date MetricBase time series data for the metric starting with this date.
        Training Dataset End Date MetricBase time series data for the metric ending with this date.
        Valid until Date that serves as a reminder to consider retraining the model. If the model is performing well, there's no need to retrain it. The model can continue working past this date.
        Active Option to use the trained model. Once the model is active it becomes available for use as a Flow Designer trigger.
      3. Click Submit and Train.
        MetricBase trains the model. When complete, the model appears on the MetricBase Model Instances tab.
      4. Click the model name.
        The modeling data appears as does the model string with the parameters optimized by the training.
        Trained model data
      5. (Optional) Click the model name and then click Set Model to change the model parameters.
        You can edit the model parameters when you want to override the settings for training your model. The graph does not update, you are saving the revised model string.
        Model parameters

      What to do next

      You can create a Flow Designer trigger for this model. For more information, see Create a model trigger.

      Tags:

      Feedback

          Share this page

          Got it! Feel free to add a comment
          To share your product suggestions, visit the Idea Portal.
          Please let us know how to improve this content

          Check any that apply

          To share your product suggestions, visit the Idea Portal.
          Confirm

          We were unable to find "Coaching" in Jakarta. Would you like to search instead?

          No Yes
          • Contact Us
          • Careers
          • Terms of Use
          • Privacy Statement
          • Sitemap
          • © ServiceNow. All rights reserved.

          Subscribe Subscribed Unsubscribe Last updated: Tags: January February March April May June July August September October November December No Results Found Versions Search preferences successfully updated My release version successfully updated My release version successfully deleted An error has occurred. Please try again later. You have been unsubscribed from all topics. You are now subscribed to and will receive notifications if any changes are made to this page. You have been unsubscribed from this content Thank you for your feedback. Form temporarily unavailable. Please try again or contact  docfeedback@servicenow.com  to submit your comments. The topic you requested does not exist in the release. You were redirected to a related topic instead. The available release versions for this topic are listed There is no specific version for this documentation. Explore products Click to go to the page. Release notes and upgrades Click to open the dropdown menu. Delete Remove No selected version Reset This field is required You are already subscribed to this topic Attach screenshot The file you uploaded exceeds the allowed file size of 20MB. Please try again with a smaller file. Please complete the reCAPTCHA step to attach a screenshot
          Log in to personalize your search results and subscribe to topics
          No, thanks Login