查看文章

Modern medicine generates a huge quantity of information daily which is stored in the medical databases. Extracting useful knowledge and providing scientific decision-making for the diagnosis and treatment of disease from the database has become a necessity. The preferred data mining functionality is association rule mining as rules are simple to understand and infer. For a rule based system to be usable in the medical domain, it must exhibit high predictive accuracy and be comprehensible. This paper surveys the various techniques for rule mining in the medical domain, identifies gaps and proposes a novel hybrid framework for efficient rule mining. A pilot study conducted over medical data paved the way for the framework. The output of the system can be used to discover new associations, validate previous findings or for the task of classification. Section I discusses the association rule mining problem. Section II discusses traditional approaches to rule mining. Section III lists the gaps in research. Section IV describes our proposed framework which includes a novel interestingness measure embedded in mining process to make it tailored to medical domain. Section V concludes the paper.