Decision theory can be used to define what constitutes an optimal decision tree, given (1)
the costs of the tests, (2) the costs of classification errors, (3) the conditional probabilities of
test results, given sequences of prior test results, and (4) the conditional probabilities of
classes, given sequences of test results. However, searching for an optimal tree is infeasible
(Pearl, 1988). ICET was designed to find a good (but not necessarily optimal) tree, where
“good” is defined as “better than the competition” (i.e., IDX, CS-ID3, and EG2).