Font Size: 

Knowledge frontier discovery is a novel technique for identifying interesting subpopulations of a dataset with respect to classification performance. A knowledge frontier is a collection of meaningful groups where any sub-partition with significantly different predictive accuracy is not meaningful. This research introduces knowledge frontiers and knowledge frontier discovery. The first knowledge frontier discovery algorithm, COBWEB-KFD, is also described in detail. Knowledge frontier discovery extends clustering and subgroup discovery to provide information that those techniques cannot.