Page 34 - MSDN Magazine, November 2017

MACHINE LEARNING
Azure Machine Learning
Time Series Analysis
for Anomaly Detection
Dawid Borycki
Anomaly Detection is one of the most important features of Internet of Things (IoT) solutions that collect and analyze tem- poral changes of data from various sensors. In many scenarios, sensor data doesn’t change significantly over time. However, when it does, it usually means that your system has encountered an anomaly—and this anomaly can lead to a specific malfunction. In this article I’ll show you how to use Azure Machine Learning Time Series Anomaly Detection to identify anomalous sensor readings. To this end I’ll extend the RemoteCamera Universal Windows Platform (UWP) app I developed in my previous article (msdn.com/ magazine/mt809116) by adding a list that displays anomalous values
(see Figure 1). The RemoteCamera app acquires images from the webcam and calculates their brightness, which fluctuates around some specific value unless the camera image changes significantly. Because you can easily induce serious brightness changes (by covering the camera, for example), leading to irregularities, this app provides good input for time-series anomaly detection.
Figure 1 Detecting Anomalous Brightness Values with Azure Machine Learning
This article discusses:
• Using Azure Machine Learning for time-series processing
• Creating and configuring machine learning models for anomaly
detection
• Publishing machine learning algorithms as Web services
• Implementing a UWP app identifying anomalies with machine learning
Technologies discussed:
Universal Windows Platform, Azure Machine Learning, REST
Code download available at:
msdn.com/magazine/1117magcode
30 msdn magazine