An Approach to Temporal-Spatial Pattern Detection

The following approach is used to mine for temporal-spatial patterns in large data sets. We rst present a parameterizable model for spatial-temporal sequences, then devise eecient ways to search for parameter values that will result in a good t to (at least portion of) the data. Metrics for how well the model ts the data are deened which are similar to the support and conndence used in association rule mining and a new metric called density is proposed. A user speciies thresholds for these metrics and, based on properties of the class of models we are attempting to t to the data, the search space can be drastically pruned. Experimental results on real and synthetic data sets demonstrate the eeciency of our algorithm.