Heuristic selection of aggregated temporal data for knowledge discovery

Abstract

We introduce techniques for heuristically ranking aggregations of data. We assume that the possible aggregations for each attribute are specified by a domain generalization graph. For temporal attributes containing dates and times, a calendar domain generalization graph is used. A generalization space is defined as the cross product of the domain generalization graphs for the attributes. Coverage filtering, direct-arc normalized correlation, and relative peak ranking are introduced for heuristically ranking the nodes in the generalization space, each of which corresponds to the original data aggregated to a specific level of granularity.