Database Reference
In-Depth Information
costs across different nodes, as well as computational, memory or storage
requirements at each node. There are several management and mining
challenges in such cases. When the streams are collected with the use of
sensors, one must take into account the limited storage, computational
power, and battery life of sensor nodes. Furthermore, since the network
may contain a very large number of sensor nodes, the effective aggrega-
tion of the streams becomes a considerable challenge. Furthermore, dis-
tributed streams also pose several challenges to mining problems, since
one must integrate the results of the mining algorithms across different
nodes. A detailed discussion of several distributed mining algorithms
areprovidedin[4].
4. Sensor Applications of Stream Mining
Data streams have numerous applications in a variety of scientific
scenarios. In this section, we will discuss different applications of data
streams and how they tie in to the techniques discussed earlier.
4.1 Military Applications
Military applications typically collect large amounts of sensor data, for
their use in a variety of applications such as the detection of events and
anomalies in the data. Some classic examples of military applications
are as follows:
4.1.1 Activity Monitoring.
Military sensors are used for a
variety of scenarios such as the detection of threats movements, sounds,
or vibrations in the underlying data. For example, the movement of en-
emy tanks in a particular region may result in a particular combination of
signals detected in the sound and activity sensors. Such monitoring may
require the development of heterogeneous mining and fusion techniques
[73], which can combine information from multiple sources in order to
perform more effective mining. Such monitoring requires stream mining
methods for the continuous detection of abnormalities, or for performing
continuous queries in the underlying data [21, 22, 26, 66, 79].
4.1.2 Event Detection.
This is related to the problem of
activity monitoring, in that specific events are captured from the stream
with the use of mining techniques. This requires the design of event
detection algorithms from data streams. This typically requires the use
of supervised learning algorithms in which the relationship of the events
to the underlying stream attributes are learned from the training data.
For example, such streams are quite common in social networks, in which
