THE UTILITY OF BACKGROUND KNOWLEDGE IN LEARNING MEDICAL DIAGNOSTIC RULES
Nada Lavrač, Sašo Džeroski, Vladimir Pirnat, Viljem Križman · Applied Artificial Intelligence · 1993
Inductive learning algorithms have frequently been applied to the problem of learning medical diagnostic rules. Most learning algorithms use an attribute-value language to describe training examples and induced rules. Consequently, the background knowledge that can be used in the learning process is of a very restricted form. To overcome these limitations, the inductive learning system LINUS incorporates attribute-value learners into a more powerful logic programming framework in which background knowledge can be used effectively. This paper describes the application of LINUS to the problem of learning rules for early diagnosis of rheumatic diseases. In addition to the attribute-value descriptions of patient data, LINUS was given background knowledge provided by a medical specialist. Medical evaluation of the rules induced by UNUS using the CN2 attribute-value learner and measurements of their performance in terms of classification accuracy and information content show that the use of background knowledge substantially improves the quality of induced rules.