Information Technology Reference
In-Depth Information
occurrences of rapid upstream movement (
um
), 6 occurrences were immediately
preceded (within two days) by a moderate water temperature (
mwt
) event, i.e., result-
ingina
con f idence
e
v
ent
(
071. Even though not in the
explicit context of movement analysis, Laube et al. (
P4
.
2008a
) discussed an adap-
tation of similar measures for spatio-temporal data mining in general. The paper
suggested spatially explicit definitions of the two measures where the classic market
basket or transactionmetaphor was replaced by spatio-temporal proximity. Crucially,
this proximity was expressed based on fuzzy concepts.
Data mining methods can produce large numbers of objectively strong and inter-
esting patterns or rules, that are however of no interest to the user. For that rea-
son, subjective measures have been suggested.
Subjective interestingness measures
depend on the class of users exploring the data, bearing in mind that patterns that are
of interest for one user class, may be of no interest to another class. Silberschatz and
Tuzhilin (
1996
) identify two reasons why a pattern is interesting from a subjective
point of view:
unexpectedness
, which indicates how surprising the pattern is to a user,
and
actionability
which indicates whether the user can act on the pattern to his/her
advantage.
E
um
→
E
mwt
)
=
6
/
84
=
0
.
3.3.3 Efficiency
Useful movement mining techniques should comply with minimal requirements in
terms of computing costs. To this end, efficiency is typically used to describe proper-
ties of an algorithm relating to how much of various types of resources it consumes.
Hence, the notion of efficiency can vary depending on which resource is of special
interest. For example, in the limited computing environments of geosensor nodes,
the number of messages required for completing a task can be a vital performance
factor. In Both et al. (
P19
.
2013
) mobile sensor nodes are tasked with monitoring
the flow in a cordon-structured transportation network. Figure
3.11
then illustrates
the scalability of three related proposed algorithms in terms of the number of mes-
sages sent when the number of fish (a.k.a. sensor nodes) in the system is increased.
The three algorithms represent three levels of complexity of decentralized commu-
nication (1. wired cordons, 2. fish carry information packages in between cordons,
3. fish also exchange information packages as they meet). The figure confirms the
expectation that algorithms 1 and 2 scale linearly, whilst algorithm 3 (added fish-fish
communication) expresses in the worst case a communication complexity that scales
with the square of the total number of fish
F
, hence
O
2
.
This topic includes a wide variety of papers, some having a more conceptual focus
whilst others have a more algorithmic character. Some conceptual papers tolerate
roughly quadratic running times, but indicate possible optimization strategies. For
instance the string-matching process used in Dodge et al. (
P14
.
2012
) is admittedly
rather slow (i.e.,
O
(
|
F
|
)
n
2
with
n
representing the number of “letters” in the trajectory).
However, pruning techniques have been used for related problems and could be
also be implemented for an operational use of the proposed concepts. By contrast,
(
)