Looking Backward, Forward, and All Around: Temporal, Spatial, and Spatio-Temporal Data Mining

May 2002

Proceedings of the Fifteenth International Florida Artificial Intelligence Research Society Conference (FLAIRS 2002)

Volume

Proceedings of the Fifteenth International Florida Artificial Intelligence Research Society Conference (FLAIRS 2002)

All Papers

We describe current research in temporal, spatial, and spatiotemporal data mining. In these types of data mining, a model of time, space, or space-time plays a nontrivial role. As an example of current research, we describe our MegaMiner prototype software. The DGG-Discover 5.2 module of MegaMiner is based on expected distribution domain generalization graphs (EDDGGs), which allow detailed domain knowledge about temporal and spatial generalization relationships to be specified, and then applied during the data mining process. As well, user expectations about the data can be specified and updated during the mining process. We illustrate the current state of the MegaMiner software by applying it to a previously unseen data set, describing the weather of the province of Saskatchewan for the period 1900 to 1949. We were able to find temporal and spatial relationships, but not spatio-temporal ones.