Make Anomaly Detection Great Again
Wednesday, September 14, 2016 - 9:00am - 9:50am
Sensors on the Internet of Things generate vast quantities of streaming data. Examples include data from wearables, home monitoring devices and traffic sensors. Fast and reliable detection of anomalies on these streams has value to people and organizations. Given the data’s speed and scale, human monitoring is impossible. Manually configured thresholds to set alarms, e.g., flag an alarm if a measurement drops below 100 ticks/minute has led to alarm fatigue. In this talk, we describe an algorithm that can automatically detect anomalies in high-speed, numeric, multi-dimensional data streams. We describe the algorithm’s performance on public data.