An efficient classifier design integrating Rough Set and Dempster-Shafer Theory

An integrated approach of knowledge discovery has been proposed in the paper using Rough Set Theory (RST) and Dempster-Shafer's (D-S) theory where high dimensional data is reduced in two folds. Firstly, unimportant attributes are eliminated using RST generating minimal subset of attributes, called reducts. Considering each core attribute as root of a decision tree, classification rules are built and grouped based on some similarity measure. Representative of each group constitute the new rule set and thus rules has been reduced while important information are retained. D-S theory ensembles the rules from which a classifier with highest accuracy has been selected.