Extracting easily interpreted diagnostic rules

Diagnosis support systems are often disregarded because of their high costs, complicated inference and inability to modify the knowledge base. The aim of this work is to propose a method that helps to resolve these problems by extracting diagnostic rules that can be easily interpreted and verified by experts. The rules can be obtained from data, even if the latter are imperfect, which is usual in medical databases. Next, intuitively clear reasoning is suggested to elaborate on the diagnosis. Rules are focal elements in the framework of the Dempster-Shafer theory. They include fuzzy sets in their premises. Thus, a measure of imprecision as a fuzzy membership function and a measure of uncertainty as the basic probability value are used. Moreover, a rule selection algorithm and a rule evaluation method that prevent some of the imperfections of the existing methods are proposed. Particular attention is paid to the evaluation of the extracted rule set according to its reliability and clarity for a human user. Experimental results obtained for popular medical data sets demonstrate the advantages of the proposed approach. For each data set, simple and readable rule sets are determined. They provide comparable or better results than the approaches published so far.