An Information Theory Approach to Diagnosis

In the preceding paper, a model of a sequential diagnostic test procedure is developed for application to fault location in electronic equipment. The average cost of diagnosis is defined and the problem of finding procedures of minimum average cost is solved for two special cases. In the present paper, the ratio of the average information gained by performing a given test to the cost of the test is introduced as a figure of merit for the test. Repeated application of this figure of merit to choose successive tests results in a systematic way of constructing efficient sequential test procedures. The procedures so constructed are compared with known optimum procedures for several special cases as a means of evaluating the information theory approach. For some special cases the true optimum is obtained; in others, the information theory test procedures are only slightly less efficient than the optimum. The advantage of the information theory approach lies in the fact that it results in a simple systematic way of constructing efficient test procedures even in the general case for which the true optimum solution has not been obtained by other means.