Information Technology Reference
In-Depth Information
condense the raw movement data by suggesting similarities and clusters or segments
and patterns. The structured data is then visualized and the human analyst can use
his/her strength of confirming the expected and detecting the unexpected by visually
inspecting the displayed information, often in an interactive analysis setting (Thomas
and Cook
2006
; Andrienko and Andrienko
2007
).
Even though the focus of the work summarized in this topic is clearly neither
visualization nor visual analytics, the section on movement mining tasks concludes
with selected examples where visual and exploratory approaches support the wider
movement mining process. These examples are:
•
Small multiples
. Small multiples (Tufte and Graves-Morris
1983
) or collections
(Bertin et al.
1981
) are a set of juxtaposed data representations allowing sliced
insight into multivariate data. Given that movement data often involves large data
sets of multiple objects with overlapping and repetitive space use, small multi-
ples are a widespread tool in movement mining. The first example included here
features small multiples for nine maps illustrating the space-use patterns of nine
GPS-tagged common brushtail possums (
P10
. Dennis et al.
2010
, Fig. 3, p. 23).
The second example illustrates for ten individual cows three different movement
parameters at six temporal scales each (
P13
. Laube and Purves
2011
, Fig. 6, p. 412).
•
Aggregation
. Laube et al. (
P3
.
2007
, Fig. 13, p. 495) features an example of aggre-
gation. Here, the spatially explicit property of flight sinuosity of a large set of trajec-
tories was interpolated into one field aggregating the information of all pigeons.
Figure
3.8
reveals that the homing pigeons show highest sinuosities around the
release site, indicating a phase of re-orientation after release.
•
Interactive interfaces
. Even though not explicitly being a result of the research sum-
marized in this volume, several interactive interfaces were developed and inten-
sively used in the movement mining process, especially for plausibility testing.
For example, the agent-based simulation environment
REPAST
served as a base
for an interactive data mining interface for the development of decentralized flock
mining algorithms in Laube et al. (
P6
.
2008b
) and Laube et al. (
P12
.
2011a
). The
interfaces allowed for the live interactive adjustment of several algorithm para-
meters during simulation runs, with linked windows showing the effects of these
adjustments in a map view as well as error plots (see Fig.
3.9
).
3.3 Evaluation
As pointed out above, the unreflected application of data mining methods can easily
lead to the discovery of meaningless patterns (Fayyad et al.
1996
). Similar care is
required for the development of data mining techniques, and hence also movement
mining techniques. Evaluating if the developedmethods are sound and produce useful
and meaningful knowledge is at the same time very important and very difficult. This
section discusses concepts that can help evaluating the quality of proposedmovement
mining techniques.
Verification
,
validation
, and
credibility
are based on the terminol-
