Database Reference
In-Depth Information
canthenbeusedtodefineafunction
φ
that measures the similarity of
two uncertain values:
φ
(
|x
i
− y
i
|
)=
Pr
(
dist
(0
,|r
(
x
i
)
− r
(
y
i
)
|
) = 0)
(7.9)
This basic similarity function is then used inside the dissimilarity func-
tion between two uncertain data values
x
and
y
, and we have
dust
(
x,y
)=
−
log(
φ
(
|
x
−
y
|
))
−
k
,where
k
=
−
log(
φ
(0)), and for entire uncertain
sequences takes the following form:
DUST
(
X,Y
)=
i
dust
(
x
i
,y
i
)
2
(7.10)
The handling of uncertainty has been isolated inside the
φ
function,
and its evaluation requires to know exactly the data distribution. In
contrast to the techniques we reviewed earlier, the DUST distance is
a real number that measures the dissimilarity between uncertain data
series. Thus, it can be used in the place of the existing distance function
in mining techniques that have been developed for certain data series.
4. Discussion
In this section, we offer some insights on the approaches and tech-
niques we described earlier. This discussion is also useful for determining
promising future research directions.
4.1 Data-Aware Network Protocols
In Section 2, we described several techniques for the ecient collec-
tion of the sensed data in a WSN. All these techniques invariably claim
considerable savings in terms of required communication messages. Ex-
periments have demonstrated savings of up to 2
3 orders of magnitude,
which is very promising news for the energy savings as well, and con-
sequently the lifetime of the WSN. However, these works have not
undertaken a careful study of how the communication savings translate
to network lifetime prolongation in real deployments.
A recent study [79] focused on exactly this problem: it examined
how DBP (similar results can be obtained for other data-driven data
acquisition techniques, as well) affected the WSN lifetime, motivated by
a real-world WSN-based application deployment in an operational road
tunnel. The performance of DBP was studied in conjunction with the
commonly-used network stack composed of CTP [42], BoX-MAC [65],
and TinyOS v2.1.1. The experimental evaluation used two settings: an
operational road tunnel, and an indoor testbed (fed with the same real
−
