Mining Diagnostic Taxonomy Using Interval-Based Similarity from Clinical Databases

Experts’ reasoning in which selects the final diagnosis from many candidates consists of hierarchical differential diagnosis. In other words, candidates gives a sophisticated hiearchical taxonomy, usally described as a tree. In this paper, the characteristics of experts’ rules are closely examined from the viewpoint of hiearchical decision steps and and a new approach to rule mining with extraction of diagnostic taxonomy from medical datasets is introduced. The key elements of this approach are calculation of the characterization set of each decision attribute (a given class) and the similarities between characterization sets. From the relations between similarities, tree-based taxonomy is obtained, which includes enough information for diagnostic rules.