Incremental Meta-Mining from Large Temporal Data Sets

With the increase in the size of datasets, data mining has become one of the most prevalent topics for research in database systems. The output from this process, the generation of rules of various types, has raised the question of how rules can be considered interesting. We argue that, in many cases, it is the meta-rule that holds the most interest. That is, given a set of known rules about a dataset, it is the confluence of rules relating to a small subset of characteristics that commonly becomes the focus of interest. In earlier work, we investigated the manner in which meta-rules, rules describing rules, could be discovered and used within a data mining system. In this paper we extend this and present an approach that enable meta-rules to be found incrementally. The emphasis of the work is on temporal data mining as we find that temporal data readily lends itself to data mining techniques, however, as can be seen from the paper, the temporal component can easily be abstracted out and the results are thus also applicable in a non-temporal domain.