Situation Assessment Using Uncertain Data
Rick Pavlik, Mark Gerken, Carl Houghton, Lisa Jesse, Rebecca J. Bussjager · AIAA Infotech@Aerospace 2010 · 2010
Intelligence analysis and exploitation technology in use today utilizes behavior patterns as a primary basis for analyzing data to identify entities, assess situations of interest, and to generate predictions of upcoming events. These patterns form the basis of generalized models used by these exploitation systems to infer meaning from incoming data streams. These models, referred to as activity models, are typically multi-dimensional, with relationships defined in the temporal, spatial, and inter-entity relationship domains. For many application domains the observation data used to create and assess these models is often uncertain as limitations in the sensor network may manifest as missing or incomplete observations or as observations with uncertainties in observation time, entity location, or entity identification. For these domains, uncertainty in the observation data may affect the structure, content, and processing of activity models. Automated support for learning activity models from historical data is further complicated by these uncertainties. In this paper, we briefly describe an approach we’ve taken that leverages probabilistic reasoning, fuzzy logic, and machine learning to support situation awareness for domains with uncertain observation data. Although designed to support uncertain data, the techniques described in this paper may also be applied to domains with more certain (known) observation data.