Harrison, Lane T.
Paffenroth, Randy Clinton
In this MQP, we focus on the development of a visualization-enabled anomaly detection system. We examine the 2011 VAST dataset challenge to efficiently generate meaningful features and apply Robust Principal Component Analysis (RPCA) to detect any data points estimated to be anomalous. This is done through an infrastructure that promotes the closing of the loop from feature generation to anomaly detection through RPCA. We enable our user to choose subsets of data through a web application and learn through visualization systems where problems are within their chosen local data slice. In this report, we explore both feature engineering techniques along with optimizing RPCA which ultimately lead to a generalized approach for detecting anomalies within a defined network architecture.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.