Framework for behavioral analytics in anomaly identification

Abstract

Behavioral Analytics (BA) relies on digital breadcrumbs to build user profiles and create clusters of entities that exhibit a large degree of similarity. The prevailing assumption is that an entity will assimilate the group behavior of the cluster it belongs to. Our understanding of BA and its application in different domains continues to evolve and is a direct result of the growing interest in Machine Learning research. When trying to detect security threats, we use BA techniques to identify anomalies, defined in this paper as deviation from the group behavior. Early research papers in this field reveal a high number of false positives where a security alert is triggered based on deviation from the cluster learned behavior but still within the norm of what the system defines as an acceptable behavior. Further, domain specific security policies tend to be narrow and inadequately represent what an entity can do. Hence, they: a) limit the amount of useful data during the learning phase; and, b) lead to violation of policy during the execution phase. In this paper, we propose a framework for future research on the role of policies and behavior security in a coalition setting with emphasis on anomaly detection and individual's deviation from group activities.