Investigating the applicability of bayesian networks to the analysis of military intelligence
Date published
2008-08-04T13:43:47Z
Free to read from
Authors
Supervisor/s
Journal Title
Journal ISSN
Volume Title
Publisher
Cranfield University
Department
Engineering Systems Department
Type
Thesis or dissertation
ISSN
Format
Citation
Abstract
Description
Intelligence failures have been attributed to an inability to correlate many small pieces
of data into a larger picture. This thesis has sought to investigate how the fusion and
analysis of uncertain or incomplete data through the use of Bayesian Belief Networks
(BBN) compares with people’s intuitive judgements. These flexible, robust, graphical
probabilistic networks are able to incorporate values from a wide range of sources
including empirical values, experimental data and subjective values. Using the latter,
elicited from a number of serving military officers, BBNs provide a logical
framework to combine each individual’s set of one-at-a-time judgements, allowing
comparisons with the same individuals’ many-at-a-time, direct intuitive judgements.
This was achieved through a serie s of fictitious and historical case studies.
Building upon this work, another area of interest was the extent to which different
elicitation techniques lead to equivalent or differing judgements. The techniques
compared were: direct ranking of the variables’ perceived importance for
discriminating between given hypotheses, likelihood ratios and conditional
probabilities. The experimental results showed that individuals were unable to
correctly manipulate the dependencies between information as evidence accumulated.
The results also showed varying beliefs about the importance of information
depending upon the elicitation technique used. Little evidence was found of a high
correlation between direct normative rankings of variables’ importance and those
obtained from the BBNs’ combination of one-at-a-time judgements. Likelihood
values should only be used as an elicitation technique by those who either regularly
manipulate uncertain information or use ratios. Overall, conditional probability
distributions provided the least troublesome elicitation technique of subjective
preferences.
In conclusion, Bayesian Belief Networks developed through the use of subjective
probability distributions offer a flexible, robust methodology for the development of a
normative model for the basis of a decision support system for the quantitative
analysis of intelligence data.
Software Description
Software Language
Github
Keywords
DOI
Rights
© Cranfield University 2008. All rights reserved. No part of this publication
may be reproduced without the written permission of the copyright owner.