Dynamic Anomaly Detection
Paragon's primary technology offering is a patent-pending dynamic anomaly detection method that can be applied to any general, time-dependent input data set. One particular strength of this new framework is that we can identify anomalous behavior without having to specify in advance what constitutes normal or abnormal behavior. This new technique seeks to answer two key questions:
- What groups of entities are behaving significantly differently than the others in the data sets?
- Have any groups of entities drastically changed their behaviors relative to the recent past?
To date, we have obtained promising results on the MIT Reality Mining cell phone data set, the Enron email data set, and Dark Web portal from the University of Arizona. We also performed tests on an email data set with ~450K emails and have obtained similarly positive results. Based on methods of complex systems and dynamical systems theory, our analysis framework can be applied to critical problems such as:
- telephone, credit card, banking, insurance, and online auction fraud prevention
- anti-money laundering
- network traffic analysis and defense against cyberattacks
- insider threat detection
- counter-terrorism and counter-narcotics data mining
- hyperspectral imagery anomaly detection
- battlespace awareness
- general information fusion
Detection of Mediated Communications
In earlier research, we developed a patent-pending counter-terrorism technology to detect the differences between the communication patterns of normal social networks and mediated communication patterns that could be indicative of covert terrorist or criminal networks. Those research results are available in this non-proprietary technical white paper.