Intelligent Probabilistic Decision System
David L. Allen, K. Wojtek Przytula, Steven B. Seida · AIAA Infotech@Aerospace 2010 · 2010
In this paper we present an intelligent decision support system based on probabilistic graphical models. We have developed a suite of tools allowing rapid model development, model verification and validation, and deployment. To support the decision maker the reasoning engine ranks decisions, uses value of information to determine the most relevant additional information to collect, and provides explanations to support recommended decisions. These tools have been applied to multiple domains, including Border protection which we present in this paper. I. Introduction ntelligent decision systems are used in a broad range of applications including autonomous systems, robotics, situational awareness, etc. They can operate autonomously or they may provide decision recommendations for a human being, who makes the final selection of an appropriate action. The decisions may be derived using deterministic reasoning, e.g. decision tree or rules, or they may require some form of uncertain reasoning, e.g. fuzzy logic or probabilistic methods. In this paper we will focus on decision support in the presence of uncertainty, specifically by applying graphical probabilistic models. In many applications decision selection is a multistep process. The initial evidence may not be sufficient for the selection of a final decision and additional evidence may need to be acquired. For example, in system failure troubleshooting, failure symptoms are often not sufficient to identify root cause failure and additional tests may have to be performed. In such multi-step process it is desirable to provide recommendations for selection of the next best item of evidence in addition to the decision selection. Another important element of decision support is explanation of the recommendation. A human decision maker may not trust the recommendation unless a convincing explanation is provided. We will present a methodology and tools for rapid development of decision support solutions, which include recommendation of additional evidence and explanation of the selected decisions. There exist several techniques for implementation of decision support in the presence of uncertainty. They include: fuzzy logic, Dempster-Shaffer and probabilistic methods. Our approach is based on graphical probabilistic models called Bayesian networks. 1