User-Driven Predictive Visual Analytics on Multivariate, Spatio-Temporal Incident Reports
J. Aurisano, EVL/UIC
Authors: Aurisano, J., Cha, M., Snyder, M.
Publication: Visualization for Predictive Analytics workshop at IEEE VisWeek, Paris, France
In this paper, we develop and implement approaches to user-driven predictive visual analytics on multivariate, spatio-temporal incident reporting data on the Lords Resistance Army (LRA) activity in Central Africa. We concentrate on specific predictive questions that pivot on LRA movement and cause-and-effect patterns. Our approach represents the output of simple models, such as a movement model and a local-related-incidents model. In addition, we provide users with interactive features to explore, which allows for the isolation of appropriate subsets of the data and visually separates signal from noise when addressing predictive questions. We implement this approach and use it to develop predictive hypotheses around variations in movement patterns of the LRA and to identify potential cause-and-effect patterns around the activity of commanders and solider defections.
Index Terms: Visual analytics, prediction, spatio-temporal data
Date: November 9, 2014
Document: View PDF