The authors developed a decision tree-critiquing program (called BUNYAN) that identifies potential modeling errors in medical decision trees. The program's critiques are based on the structure of a decision problem, obtained from an abstract description specifying only the basic semantic categories of the model's components. A taxonomy of node and branch types supplies the primitive building blocks for representing decision trees. BUNYAN detects potential problems in a model by matching general pattern expressions that refer to these primitives. A small set of general principles justifies critiquing rules that detect four categories of potential structural problems: impossible strategies, dominated strategies, unaccountable violations of symmetry, and omission of apparently reasonable strategies. Although critiquing based on structure alone has clear limitations, principled structural analysis constitutes the core of a methodology for reasoning about decision models. Key words: decision trees; computer-assisted critiquing. (Med Decis Making 1989;9:272-284)
[1]
I. Good.
On the Principle of Total Evidence
,
1967
.
[2]
Ross D. Shachter.
Evaluating Influence Diagrams
,
1986,
Oper. Res..
[3]
Peter Szolovits,et al.
Artificial intelligence in medical diagnosis.
,
1988,
Annals of internal medicine.
[4]
Lawrence M. Fagan,et al.
A Methodology for Generating Computer-based Explanations of Decision-theoretic Advice
,
1988,
Medical decision making : an international journal of the Society for Medical Decision Making.
[5]
L. Booker,et al.
Integrating logical and probabilistic reasoning for decision making.
,
1988
.
[6]
Michael P. Wellman.
Qualitative Probabilistic Networks for Planning Under Uncertainty
,
1990,
UAI 1990.
[7]
Michael P. Wellman.
Fundamental Concepts of Qualitative Probabilistic Networks
,
1990,
Artif. Intell..