Anomaly Detection

---

MAC's Data Exploitation Group provides customizable software and services for anomaly and novelty detection. Products for discovering novel events and detecting anomalous behavior are quickly becoming indispensable for the proper operation and maintenance of the complex systems employed by modern industry, medical providers and the military. Factories, health monitors, and aircraft and other vehicles regularly produce hundreds or thousands of channels of real-time telemetry which must be monitored for possible indications of failure. These data resources present an enormous opportunity for software tools that can detect events of interest. These tools can be used to discover valuable new insights into operations, diagnose faults post-mortem to determine the cause, and predict failures before they cause costly damage or injury.

Our group has developed powerful pattern recognition technology, based on Ensembles of Decision Trees (EDTs), which drives our tools for novelty and anomaly detection. EDT-based technology has many advantages over existing techniques, including the scalability to handle very high dimensional data, automatic drill-down facilities to explain why an event is categorized as novel or anomalous, and extremely rapid evaluation. The approach is also amenable to deployment on FPGA devices, Graphical Processing Units, parallel processors and other accelerated platforms.

This figure illustrates MAC’s approach to Anomaly Detection from the input from various types of monitoring devices through the processing by Ensembles of Decision Trees which yields a signal indicating the location of anomalies or unusual combinations of signals along with a ranked list of the input measurands that explain why that interval seems to be noteworthy