CSS SEMINAR - Nathan Danneman

Friday, January 31, 3:00 p.m.
Center for Social Complexity Suite
Research Hall, Third Floor

Identifying Potential Smugglers: Local (Geospatial) Outlier Detection

Nathan Danneman, PhD
Data Tactics

ABSTRACT: Multivariate outlier detection is a common, yet challenging, inferential task. This talk presents a novel composite model, specifically, a local, unsupervised-as-supervised, bootstrap-aggregated probability model, to detect outliers in multivariate data. The method, though highly general, does not free users from making important choices about aggregation and model flexibility. The method turns out to be useful for geospatial outlier detection, as demonstrated in a use case involving potential smugglers in the Strait of Hormuz.

Nathan Danneman earned his PhD in political science from Emory University in 2013, where he studied international conflict through a combination of game theory and computational statistics. He currently works at Data Tactics where he supports various DHS and DARPA projects. His current research efforts focus on outlier detection, as well as the analysis of text data.