Database Reference
In-Depth Information
Query Processing Stage
and association mining approach for stream data
application based on multiple sensor relationships.
We compared the estimation accuracy, running
time and memory space usage when applying
each method to our proposed framework.
At the query processing stage, which is on the
top level of the domain-driven framework, dif-
ferent users' query can be fulfilled at the users'
specified query criteria at the same time. The end
user will specify their query parameters on their
online queries, in this case, the support, confi-
dence threshold, and the sliding window size of
their online association queries, or specify they
would use the system recommend parameter val-
ues. The query processor then use these specified
parameters as query criteria to pass the request
to the data warehousing and mining components
and retrieve the query results for different end
users' requests.
Performance Study of
Estimation Accuracy
The evaluation of the estimation accuracy of the
missing values is done by using the average Root
Mean Square Error (RMSE):
#
estimations
å
2
(
Xa
-
Xe
)
1
i
i
RMSE
=
i
=
1
numStates
#
estimations
Performance Study
where X
ai
and X
ei
are the actual value and the
estimated value, respectively; #estimations is the
number of estimations performed in a simulation
run; and numStates is the number of states in which
the actual readings are distributed.
The performance of our proposed framework is
studied by means of simulation. Several different
data warehousing and data mining experiments
are conducted in order to evaluate the proposed
framework while using the Average Window
Size (AWS) approach, the linear interpolation
approach, the linear trend approach, the Window
based Association Rule Mining (WARM) ap-
proach (Halatchev, 2005), and the Closed itemsets
basedAssociation Rule Mining (CARM) approach
(Jiang, 2007).
All these methods are applied to our proposed
framework to answer the user's request for missing
sensor value estimation. The AWS approach uses
the average value of the missing sensor readings
in the current sliding window as the estimated
value. The linear interpolation approach uses
the linear interpolation of the missing sensor's
neighbor readings as the estimated value. The
linear trend uses the linear regression trend for the
missing sensor readings as the estimated value. The
WARM approach (Halatchev, 2005) is a revised
data warehousing technique for sensor network
database based on the relationship between two
sensors. And the CARM approach is a pattern
#
estimations
å
2
(
Xa
-
Xe
)
i
i
The expression
repre-
i
=
1
#
estimations
sents the standard error and is an estimate of the
standard deviation under the assumption that the
errors in the estimated values (i.e. X
ai
- X
ei
) are
normally distributed. Thus, the RMSE allows the
construction of confidence intervals describing the
performance of different candidate missing value
estimators. The smaller the RMSE (the standard
deviation), the better the estimated results.
From Figure 5, we can see that CARM gives
the best result of the above approaches in the
proposed framework regarding the estimation
accuracy, followed by the WARM approach.
The linear interpolation, AWS, and linear trend
approaches perform no better than WARM and
CARM approaches. The main reason might be
that it only considers the relationship between the
neighbor nodes, while CARM and WARM find
Search WWH ::
Custom Search