Assessing the Robustness of Graph Statistics for Network Analysis Under Incomplete Information
Abstract
Due to the emergence of powerful global terrorist organizations such as Al Qaeda and ISIS over the last 15 years, social network analysis is increasingly leveraged by the Department of Defense to develop strategies to combat criminal and terrorist organizations. Understanding and correctly classifying networks improves our ability to destroy criminal and terrorist networks because we can leverage existing literature that identifies the optimal strategy for dismantling these networks based on their network structure. However, these strategies typically assume complete information about the underlying network. Due to the limited ability of an analyst to process all of the available data, our inability to detect all members of these networks, and the efforts of criminal organizations to hide their activities and structure, analysts must classify these networks and develop strategies to combat them with missing information. This thesis analyzes the performance of a variety of network statistics in the context of incomplete information by leveraging simulation to remove nodes and edges from networks and evaluating the effect this missing information has on our ability to accurately classify the underlying structure of the network. We provide recommendations to intelligence analysts about which statistics provide the most information, conditions under which it is reasonable to assert a classification, and a framework for the evaluation of network statistics for the purposes of classifying network graphs under incomplete information.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2018
- Accession Number
- AD1052532
Entities
People
- Xian L. Chia
Organizations
- Naval Postgraduate School