Multi-Entity Bayesian Networks learning for hybrid variables in situation awareness

Over the past two decades, machine learning has led to substantial changes in Data Fusion Systems throughout the world. One of the most important application areas for data fusion is situation awareness. Situation Awareness is perception of elements in the environment, comprehension of the current situation, and projection of future status before decision making. Traditional fusion systems focus on lower levels of the JDL hierarchy, leaving higher-level fusion and situation awareness largely to unaided human judgment. This becomes untenable in today's increasingly data-rich environments, characterized by information and cognitive overload. Higher-level fusion to support situation awareness requires semantically rich representations amenable to automated processing. Multi-Entity Bayesian Networks (MEBN) combine First-Order Logic with Bayesian Networks for representing and reasoning about uncertainty in complex, knowledge-rich domains. MEBN goes beyond standard Bayesian networks to enable reasoning about an unknown number of entities interacting with each other in various types of relationships, a key requirement for PSAW. A MEBN model can be constructed manually by a domain expert or automatically by a machine learning algorithm. A discrete MEBN learning algorithm was recently developed. However, many real world variables are continuous. This paper presents a hybrid (both discrete and continuous variables) MEBN learning algorithm. The method is evaluated on a case study from the PROGNOS, predictive situation awareness system.