A probabilistic approach for reducing the search cost in binary decision trees

Abstract

A probabilistic model for reducing the number of decisions (tests) that are required in a particular diagnostic procedure is presented. Specifically, it is considered that a problem is structured as a binary balanced decision tree the interior nodes of which represent test points; the paths of the three correspond to different diagnoses. By assuming that there is sufficient probabilistic information available concerning the decisions at the interior nodes, attempt is made to minimize the average number of these decisions when one searches for a final diagnosis. A gain function is built up and the expression for its parameters is derived. Two heuristic methods are proposed for the selection of the nodes where a decision are proposed for the selection of the nodes are compared in terms of the value of the gain achieved. >