Spatiotemporal Change Pattern Mining: A Multi-disciplinary Perspective - Space-Time Integration in Geography and GIScience - page 314

Table 17.2 List of techniques and related categories

Pattern name
used

Change pattern
footprint

Primary
applications Techniques Area of research

CUSUM Time series analysis Abrupt change,

change point

Time point System control

S-Outlier detection Spatial statistics Spatial outlier Spatial location Image analysis,

transportation

Interesting interval
discovery

Time series
analysis, spatial
statistics

Sub-
path/interval of
abrupt change

Interval Climate science

Statistical spatial
Wombling

Spatial statistics Wombling, rapid

change, abrupt
change

Spatial curves Public health,

ecology

Scan Statistics Spatial statistics,

cluster detection

Spatial cluster Spatial regions Public health

Change detection Image Processing Change Spatial

regions/pixels

Remote sensing

ST cluster
detection

Spatial statistics ST cluster,

outbreak

ST subspace Public health

Emerging ST
cluster detection

Spatial statistics Emerging cluster ST subspace Public health

keeps a cumulative sum of a score S 0 based on a statistical parameter ™ of the
data stream. When the sum exceeds a certain threshold, a change can be flagged.
Formally, the score (for detecting positive change) is defined as S 0 D 0, and
S n C 1 D max(0, S n C x n n )where x n is the nth data value, and n is the parameter.
A number of works under the CUSUM framework have been proposed, defining
as the mean, standard deviation, or likelihood function of the data (Kawahara and
Sugiyama 2009 ; Kucera et al. 2007 ).
Other change point detection techniques may use likelihood ratio tests,
maximum-likelihood estimation (MLE), and model learning to find the change
point. Detailed reviews of techniques can be found in the time series analysis
literature (Basseville and Nikiforov 1993 ; Shaban 1980 ; Zacks 1983 ).

17.5.2 Spatial Outlier Detection Techniques

Spatial outlier detection techniques have been discussed in the literature of spatial
statistics. One example is the Moran Scatterplot (Anselin 1995 ) approach. The z-
score (normalized value) at a location is plotted against the average z-score at its
neighboring locations. Points falling into the second and the fourth quadrants are
considered as spatial outliers. Other statistical approaches include variogram cloud,
scatterplot, and spatial statistic Z s(x) , etc. A unified definition of spatial outlier
detection along with a computational framework can be found in the literature of
data mining (Shekhar et al. 2003 ).