Category and Partition Utility

Publication	Fisher/87a: Knowledge Acquisition Via Incremental Conceptual Clustering
Name	Category and Partition Utility
Description	In order to achieve high predictability of variable values, given a cluster, and high predictiveness of a cluster, given variable values, the clustering algorithm COBWEB measures the utility of a cluster (Clustering Utility) as: CU(Ck) = P(Ck) * ∑i∑j [P(Ai = Vij|Ck)2 - P(Ai = Vij)2] The utility of a partition of data (Partition Utility) is defined as: PU({C1, ..., CN}) = ∑kCU(Ck) / N The aim of COBWEB is to hierarchically cluster the given observations (unclasssified examples) in such a way, that Partition Utility is maximized at each level. Please refer to the COBWEB-Link above for a more detailed description.
Algorithm	COBWEB