Induction of medical knowledge based on rough sets and cross-validation method

S. Tsumoto, Hiroshi Tanaka

Research output: Contribution to journalArticlepeer-review


Several symbolic methods of inductive learning have been proposed such as decision trees and the AQ family. These methods are applied to extract meaningful knowledge from a large database, and their application is in some aspects appropriate. However, in most cases, the methods are of deterministic nature, and the reliability of acquired knowledge is not strictly evaluated from a statistical point of view. These methods are therefore not suitable to apply to a domain of essentially probabilistic nature like that of medicine. By extending the concepts of Pawlak's rough set theory to the probabilistic domain, we introduce a new approach to knowledge acquisition called PRIMEROSE that induces probabilistic rules. We also describe a program that extracts rules from a clinical database for use with an expert system by using this method. The program is applied to three medical domains: headache, meningoencephalitis, and cerebrovascular diseases. The results show that the derived rules almost correspond to those of medical experts.

Original languageEnglish
Pages (from-to)91-99
Number of pages9
JournalJapan Journal of Medical Informatics
Issue number2
Publication statusPublished - 1996 Jan 1


  • cross validation
  • inductive learning
  • knowledge acquisition
  • medical knowledge
  • rough set theory

ASJC Scopus subject areas

  • Health Informatics


Dive into the research topics of 'Induction of medical knowledge based on rough sets and cross-validation method'. Together they form a unique fingerprint.

Cite this