Geography Reference
In-Depth Information
17.4.4
ST Subspace of Change
A ST subspace of change refers to the footprint of a change of phenomena (e.g.,
disease outbreak) occurring in both space and time. We further divide it into two
types, namely, ST clusters and ST emerging clusters.
17.4.4.1
ST Clusters
The spatial cluster pattern may be generalized to the temporal dimension. In a ST
event report dataset, a ST cluster refers to (region, time-interval) tuples where the
density of points is higher than the rest of the data, or compared to the expected
number. A ST cluster implies that a spatial region has a higher risk of disease or
crime during a certain period of time. On an aggregated or raster ST dataset, a ST
hotspot pattern can be defined in a similar manner.
17.4.4.2
Emerging ST Clusters
An emerging ST clusters is a special type of ST cluster, where the magnitude of
change (e.g., difference from expected value) is monotonically changing over time.
Though they have the same ST footprint as ordinary ST clusters, emerging clusters
characterize the evolution of ST clusters over time. They thus resemble change
interval on the temporal dimension.
A spatiotemporal version of scan statistics (Kulldorff et al.
1998
) has been
proposed to discover ST clusters. A model, proposed by Neill et al. (
2005
)employs
similar ideas for discovering emerging ST clusters.
17.5
Change Pattern Mining Approaches
This section introduces a few representative techniques from various change pattern
categories. Table
17.2
listed these techniques with their respective pattern footprint.
Due to space limitations, readers are directed to more detailed reviews of further
explorations.
17.5.1
Change Point Detection Techniques
Many change point detection techniques have been proposed. One is the Cumulative
Sum (CUSUM), first proposed by Page in 1954 (Page
1954
), CUSUM is an on-line
approach to process time series such as system monitoring signals. The method
Search WWH ::
Custom Search