Stream Analytics Change Point Anomaly Detection

116dreamer 26 Reputation points
2022-10-26T02:16:08.09+00:00

Hello! I have been attempting to use the AnomalyDetection_ChangePoint function in Azure Stream Analytics but have not gotten any results. I am generating simulated data through code to match the trends shared in the Microsoft documentation page (picture attached) however, the function is not detecting anything and I have no results. I have attached a screenshot of how my data looks and the corresponding lack of results from the Change Point function.

I would greatly appreciate any help if anyone could explain what is the issue. Thank you!
254049-microsoft.png

254161-data.png

Azure Stream Analytics
Azure Stream Analytics
An Azure real-time analytics service designed for mission-critical workloads.
330 questions
Azure AI services
Azure AI services
A group of Azure services, SDKs, and APIs designed to make apps more intelligent, engaging, and discoverable.
2,375 questions

2 answers

Sort by: Most helpful
Most helpful Newest Oldest
  1. Sid Ramadoss 161 Reputation points
    2022-10-31T20:27:06.673+00:00

    Hi @116dreamer
    If you are trying to build a working PoC/demo, I recommend using the data generator for this tutorial.
    https://learn.microsoft.com/en-us/azure/stream-analytics/stream-analytics-machine-learning-anomaly-detection#model-behavior

    I suspect there might be some problems with the number of inputs your example generates and how you have invoked the AD functions in your query. This documentation page shows a working example that you can experiment with: https://learn.microsoft.com/en-us/azure/stream-analytics/stream-analytics-machine-learning-anomaly-detection#model-behavior

    1 person found this answer helpful.
  2. Sander van de Velde 28,386 Reputation points MVP
    2022-10-28T23:14:45.747+00:00

    Hello @116dreamer ,

    Thanks for sharing the straightforward queries.

    Does it seem you expect 5 minutes of data, 1 message per second (300 data points)?

    The anomaly detection has to learn what normal input telemetry looks like for five minutes. Are you waiting five minutes before you start sending the anomalies?

    Can you also lower the confidence percentage (eg. 90 percent) so we do not have high expectations?