The use of induction in the design of an expert system for thyroid function studies

In this study, the role of induction in the design of an expert system for diagnosing thyroid disorders is evaluated. An expert system was first designed conventionally, based on interaction between a knowledge engineer and a thyroid specialist. This involved weighting three tests (T3, free T4 and TSH) according to the reliability of the test and the presence or otherwise of influencing factors. Compatibility was then tested with known ranges for the parameter values, and a diagnosis made of one of three possible outcomes (euthyroid; hypothyroid; hyperthyroid). Two expert systems were then induced using, firstly, a set of rules and, secondly, a sample set chosen from a database. These systems were then tested against the expert designed system. Both induced systems produced results which were superior to the expert designed system and, in addition, provided insight into the decision-making process. It is concluded that induction is very useful in the design of expert systems of this nature.