With the rise of the IoT and thus an increase in devices connected to the internet, large, industrial-sized users of the technology delayed the implementation of anomaly detection systems due in no small part to the need for highly customized solutions. However, with a new Azure Stream Analytics capability - which is in private preview - Microsoft seeks to extend a helping hand.
The new addition to the Stream Analytics toolset allows for the detection of anomalies in the data stream, through tracking of KPIs - Key Performance Indicators - usage monitoring, and performance monitoring - CPU usage, memory usage, etc.
Aimed more towards numerical time series data, Azure Stream Analytics can "detect positive and negative trends, and changes in a dynamic range of value" and as such be used to create alerts in case of high resource usage or "service health instability".
With the use of Machine Learning, the service can detect input data anomalies, track changes and report them as anomaly scores and binary spike indicators for every individual point in time. The underlying model here is one which uses continuous learning to improve over time, and the enabling of it is done in "a declarative SQL like query language to reason about data in motion."
Those interested in the private preview can sign up here.