Evolutionary Hot Spots Data Mining - An Architecture for Exploring for Interesting Discoveries

Data Mining delivers novel and useful knowledge from very large collections of data. The task is often characterised as identifying key areas within a very large dataset which have some importance or are otherwise interesting to the data owners. We call this hot spots data mining. Data mining projects usually begin with ill-defined goals expressed vaguely in terms of making interesting discoveries. The actual goals are refined and clarified as the process proceeds. Data mining is an exploratory process where the goals may change and such changes may impact the data space being explored. In this paper we introduce an approach to data mining where the development of the goal itself is part of the problem solving process. We propose an evolutionary approach to hot spots data mining where both the measure of interestingness and the descriptions of groups in the data are evolved under the influence of a user guiding the system towards significant discoveries.