Preliminary Content

Acknowledgements

I thank my Advisor, Professor Peter Hoff, and the Director of Undergraduate Studies, Professor Mine Cetinkaya-Rundel, for their guidance in this project. I also thank Duke University’s Statistics Department and Office of Information Technology, especially my dataset contact at OIT, Eric Hope, for making this project possible. Most of all I thank my parents for their continued unwavering support in all my endeavors.

Abstract

The goal of this project is to identify novel methods for detecting anomalies in network IP data. The space is represented as a 3-dimensional tensor of the continuous features (source bytes, destination bytes, source packets, destination packets) divided by their respective source port and destination port combinations. This project implements and assesses the validity of principal component analysis and matrix completion via singular value decomposition (more methods pending) in determining anomalous entries in the tensor.