This research is part of a proposed shift in emphasis in decision support from optimality to robustness. Computer simulation is emerging as a useful tool in planning courses of action (COAs). Simulations require domain models, but there is an inevitable gap between models and reality - some aspects of reality are not represented at all, and what is represented may contain errors. As models are aggregated from multiple sources, the decision maker is further insulated from even an awareness of model weaknesses. To realize the full power of computer simluations to support decision making, decision support systems should support the planner in exporing the robustness of COAs in the face of potential weaknesses in simulation models.
This paper demonstrates a method of exploring the robustness of a COA with respect to specific model assumptions about whose accuracy the decision maker might have concerns. The domain is that of peacekeeping in a country where three differenct demographic groups co-exist in tension. An external peacekeeping force strives to achieve stability, an improved economy, and a higher degree of democracy in the country. A proposed COA for such a force is simluated multiple times while varying the assumptions. A visual data analysis tool is used to explore COA robustness. The aim is to help the decision maker choose a COA that is likely to be successful even in the face of potential errors in the assumptions in the models.
We start with a vision of an integrated decision architecture to assist in the various stages and subtasks of decisionmaking.
We briefly describe how the Seeker-Filter-Viewer (S-F-V) architecture for multi-criterial decision support
helps realize many components of that vision. The rest of the paper is devoted to one of the components: developing
insights about the course of action (COA) decision space from COA simulations. We start with data obtained from
multiple simulation executions of an urban combat COA in a specified scenario, where the stochastic nature of different
executions produce a range of intermediate events and final outcomes. The Viewer in the S-F-V decision architecture is
used to make and visually test hypotheses about how sensitive different events and outcomes are to different aspects of
the COA and to various intermediate events. The analyst engages in a cycle of hypothesis making, visually evaluating
the hypothesis, and making further hypotheses. A set of snapshots illustrates an investigational sequence of abstractions
in an example of iterating on hypotheses. The synergy of data mining tools, high performance computing, and advanced
high-resolution combat simulation has the potential to assist battle planners to make better decisions for imminent