Detecting Threatening Behavior Using Bayesian Networks




AlGhamdi, Ghazi
Laskey, Kathryn B.
Wang, Xun
Barbará, Daniel
Shackelford, Thomas
Wright, Edward J.
Fitzgerald, Julie

Journal Title

Journal ISSN

Volume Title



This paper presents an innovative use of human behavior models for detecting insider threats to information systems. While most work in information security concerns detecting and responding to intruders, violations of system security policy by authorized computer users present a major threat to information security. A promising approach to detection and response is to model behavior of normal users and threats, and apply sophisticated inference methods to detect patterns of behavior that deviate from normal behavior in ways suggesting a possible security threat. This paper presents an approach, based on multi-entity Bayesian networks, to modeling user queries and detecting situations in which users in sensitive positions may be accessing documents outside their assigned areas of responsibility. Such unusual access patterns might be characteristic of users attempting illegal activities such as disclosure of classified information. We present a scalable proof of concept behavior model, provide an experimental demonstration of its ability to detect unusual access patterns in simulated situations, and describe future plans to increase the realism and fidelity of the model.


The views, opinions, and findings contained in this paper are those of the author(s) and should not be construed as an official position, policy, or decision, of ARDA, the Department of the Interior, or the US Navy unless so designated by other official documentation


Information security, Behavioral model, Multi-entity Bayesian networks, Document relevance, Insider threat detection, Access control