Diagnostic Rules Discovery with Hierarchical Clustering and Focusing Mechanism Based on Rough Sets Theory

An approach is proposed to discover diagnostic rules from clinical databases. First, the diseases in the clinical database are clustered by their necessary characterization. Then focusing mechanism, which includes three processes: exclusion process, discrimination process and combining process, is exploited to derive diagnostic rules. The main characteristic feature of the approach is: 1) coverage is exploited to find necessary characterization of diseases during the exclusion process, while accuracy is exploited to find lambda_A-sufficient characterization of diseases during the discrimination process; 2) discrimination process can be executed among many diseases; 3) a series of classification information systems (CISs) derived by exclusion process from the original are considered; 4) the CISs are reducted to the simple ones; 5) crisp rules and uncertain rules can be conveniently derived. Finally, an example illustrates the approach and shows its effectiveness.