The basic probability assignment as a measure of diagnostic rules significance

Diagnostic rules are usually IF-THEN rules, but they should satisfy specific requirements of a diagnosis. Thus, not always the classical methods of rules determination are applicable. In the present paper it is suggested to find out the set of rules by an elimination of superfluous rules from the maximal rule set or adding rules that improve inference to the minimal set of rules. It is shown that the basic probability assignment determined in the Dempster-Shafer theory of evidence can be used as a measure indicating symptoms that are the most significant for a diagnosis and should create rules. A set of IF-THEN rules with fuzzy premises and crisp conclusions can be built in this way. The proposed method is illustrated by determining rules allowing for diagnostic inference for a database of thyroid gland diseases.