Multivariate Methods and Data Integration in Social Media for Anomaly Detection

• Activity Number: 630 - Uncertainty Quantification, Reliability and Robust Inference
• Type: Contributed
• Date/Time: Thursday, August 2, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Defense and National Security
• Author(s): Karl Pazdernik and Kellie MacPhee and Bryan Stanfill and Lisa Bramer*
• Companies: Pacific Northwest National Laboratory and University of Washington and Pacific Northwest National Laboratory and Pacific Northwest National Laboratory
• Keywords: anomaly detection; multivariate; principal component analysis; robust; streaming; tensor

Abstract

Social media analytics have witnessed an explosion in the literature due to the growing popularity of social media platforms such as Twitter, Facebook, and Instagram. In particular, social media has been documented as a useful resource in problems relevant to national security, such as health monitoring, crime, and counter-terrorism. In this presentation, we focus on the specific statistical task of anomaly detection in multiple sources of streaming data. We compare a variety of anomaly detection techniques and discuss the impact of utilizing the hierarchical structure of temporal and multivariate data. Finally, we show how this methodology can be used for anomaly detection problems relevant to national security.