A Framework for Data Warehousing and Mining in Sensor Stream Application Domains - Evolving Application Domains of Data Warehousing and Mining - page 126

Figure 8. Performance study of average estima-
tion accuracy

Figure 10. Performance study of memory usage

multiple sensor readings of the missing senor and
the other related sensors.

between the existing sensors. The linear interpola-
tion method fails to report the missing temperature
value, because it predicates the missing value
based on the missing readings neighbor reported
values. Therefore, when there are two consecutive
missing values on the same sensor, this method
couldn't calculate the estimated result.

Performance Study of Memory Usage

Figure 10 illustrates the memory usage of AWS,
linear interpolation, linear trend, and CARM ap-
proaches in MB. The experimental results show
that in terms of memory space, the CARM ap-
proach is outperformed by all other three approach-
es. This is because CARM stores the relationships
of multiple sensor readings, and for the AWS,
linear interpolation, and linear trend approach,
they only need to store and process information
regarding the missing sensor's readings.

Performance Study of Running Time

Figure 9 illustrates the running time in seconds of
AWS, linear interpolation, linear trend, and CARM
approaches. The experimental results show that
in terms of running time, the CARM approach
are outperformed by AWS, linear interpolation
and linear trend approaches. This is because in
the AWS, linear interpolation, linear trend ap-
proaches, the calculation is based on relationships
in a certain sensors reading trend, while for CARM
the calculation is based on relationships between

Analysis Summary

In this section, we discussed how to setup and
implement an environmental monitoring sensor
stream application in the domain-driven frame-
work for data warehousing and mining from
sensor streams. We described the process of the
framework setup in each implementing stage:
the raw data stage, data preprocessing stage, data
warehousing and data mining stage, and the query
processing stage.
The performance of our proposed framework is
studied by means of simulation. Several different
data mining techniques are conducted in order to
evaluate the proposed framework using theAverage

Figure 9. Performance study of running time