Announcing New Features in Multivariate Anomaly Detector
Published Jan 26 2022 06:43 AM 4,480 Views

Microsoft partners and customers have been building metrics monitoring solutions for generic scenarios like predictive maintenance, by leveraging the easy-to-use time-series anomaly detection service in Microsoft Azure:  Multivariate Anomaly Detection, which is in public preview in last April. This API service helps more customers to proactively protect their complex systems in industries including energy, manufacturing, etc. 


What's new in this version?

We are excited to announce that we are adding more powerful capabilities in Microsoft Azure Multivariate Anomaly Detector (MVAD) today. In the latest version(v1.1-preview.1) of this API, we implemented a new , in a synchronous manner, which means you could get the anomaly detection results immediately once you call this API. This synchronous inference API is a substantial change compared with previous inference process and will be more intuitive and easier-to-use.


Also, we added a new field named 'interpretation'  to give more explanations on an anomaly, like which variables have huge correlation changes and cause the anomaly. These updates will support you to better leverage MVAD and get more useful information to analyze and take actions.


Overview of this new API

With the synchronous API, you can get inference results point by point in real time, without the need for compressing and uploading task like training and asynchronous inference. Here are some requirements for the synchronous API:

  • Need to put data in JSON format into the API request body.
  • The inference results are limited to up to 10 data points, which means you could detect 1 to 10 timestamps with one synchronous API call.
  • Due to payload limitation, the size of inference data in the request body is limited, which support at most 300 sensors’ data and 2880 timestamps.

In total, there are 8 APIs available for you to use in Multivariate Anomaly Detection, check them out in  Anomaly Detector v1.1-preview.1 API Reference.


Overview of 'interpretation'

You will get a JSON response of inference results in real time after you call a synchronous API, which contains the following new fields, and these fields will help you with root cause analysis. Therefore, you could not only know when there's an anomaly, but also know which variables contributed to this anomaly, which will save your time of exploration on all variables.


New Fields



This field only appears when a timestamp is detected as anomalous, which contains variables, contributionScore, correlationChanges.


This field only appears when a timestamp is detected as anomalous, which included in interpretation. It contains changedVariables and changedValues that interpret which correlations between variables changed.


This field will show which interrelated variables cause the anomaly.


This field indicates how much the correlation changed between variables. The bigger the number is, the greater the change on correlations.


Getting started is really simple

You have two quick options to get started:
Option 1: Follow the steps in How to: Use Multivariate Anomaly Detector on your time series data, it contains all the process for you to train a model and do inference, both in asynchronous and synchronous ways, perfectly match the need of those who are new to Microsoft Azure Multivariate Anomaly Detector and want to get some nitty-gritties of our service. This document also includes detailed codes and best practices.
Option 2: For those who want to hands on directly on code and familiar with SDK, this MVAD SDK Sample Notebook is perfect for you!


Learn More


Version history
Last update:
‎Jan 27 2022 02:25 AM
Updated by: