Here is today's citation of interesting applied research with Bayesian networks:
Stech, Frank J., and Christopher Elsaesser.
Midway Revisited: Detecting Deception by Analysis of Competing Hypotheses.
Military Operations Research 12, no. 1 (November 11, 2007): 35–55.
Historical accounts of military deception abound, but there are few historical accounts of counter-deception, and fewer operational theories. This paper describes a business process and semi-automated tools for detecting deception. Our prototype counter-deception business process starts with hypothesis generation. For tactical situations, this consists of automated course of action generation. Strategic situations rely on elicitation from analysts. Next, a Bayesian belief network is generated. This is followed by sensitivity analysis based on Bayesian classification. The result is a weighted list of possible observations that: (1) identify distinguishing evidence that a deceiver must hide and a counterdeceiver must uncover, (2) isolate local deception in intelligence reporting and sensing from global deception, and (3) identify circumstances when it might be fruitful to entertain additional hypotheses. We illustrate this model by describing how it could have been used by the Japanese Navy before the Battle of Midway to detect the American denial and deception tactics that allowed the U.S. Pacific Fleet aircraft carriers to ambush and sink four Japanese carriers.