Discovering generalized symbolic rules from un-annotated data: A hybrid of rough sets and attribute oriented induction

Rule induction is a data mining process for acquiring knowledge in terms of symbolic decision rules from a number of specific 'examples' to explain the inherent causal relationship between conditional factors and a given decision/outcome. In this thesis, we have developed a Decision Rule Acquisition Workbench (DRAW) that discovers conjunctive normal form decision rules from un-annotated data sets. Our rule-induction strategy uses (i) conceptual clustering to cluster and generate a conceptual hierarchy of the data set; (ii) rough sets based rule induction algorithm to generate decision rules from the emergent data clusters; and (iii) attribute oriented induction to generalize the derived decision rules to yield high-level decision rules and a minimal rule-set size. We evaluate DRAW with five standard data sets. Finally, we apply DRAW to derive decision rules for classifying confocal scanning laser tomography images of the optic disk for the diagnosis of glaucoma. (Abstract shortened by UMI.)...