A Study on the Choquet-Based Criteria in Possibilistic Sequential Decision Making.

This paper focuses on the optimal strategy of possibilistic decision trees with the possibilistic Choquet integrals criteria. It shows that the criteria don't satisfy monotonicity, so that the optimal solution cannot be computed in polynomial time by Dynamic Programming. Therefore, a Branch and Bound approach is developed to get the best solution. Experiments show that the strategy of Branch and Bound will be too costly to accept when the size of decision tree get a bit bigger. However, the Dynamic Programming approach is generally quite practicable and meaningful, even not always optimal. [ABSTRACT FROM AUTHOR]