Abstract. In this paper, a variety of targeted sensitivity analysis approaches
are explored for a Bayesian Belief Network (BBN) constructed as
an expert tool for enemy course of action (COA) assessment
at the tactical level in a conventional mid-intensity scenario. Robustness
analysis is used to measure the level to which the
posterior probability of the states at the root node are
affected by instantiation of individual nodes in the network. Likewize,
value of information analysis and gain in belief updating are
used to compare how nodes of interest affect posterior probabilities
at the root node, the former measuring Shannon Entropy and
the latter Kullback Distance. Finally, sensor effectiveness analysis is used
to measure how the reliability of reconnaissance and surveillance (R&S)
assets affects updating of belief at the root node. It
was found that each of the sensitivity analysis approaches could
be used to optimize allocation of R&S, to identify the
commander’s decision points, and to identify influential nodes for which
the conditional probability tables (CPTs) should be refined. In terms
of utility, it was concluded that, as in the case
of the use of BBNs in the tactical COA assessment
domain in general, the utility of sensitivity analysis of the
BBN would be reduced in conditions of high operational tempo
and myriad variables influencing tactical COA selection. Nevertheless, in a
slower operational tempo environment, the benefits in refinement and utility
of the BBN derived through sensitivity analysis would be significant.
Related topics:
Operations research, training and analysis
View first page of "Brosnan: Sensitivity Analysis of a Bayesian Belief Network in a Tactical Intelligence Application"
Papers by Brosnan
Register for the free
to receive a list of papers for each issue as it is released.
|
|
