Database Reference
In-Depth Information
1. Introduction
A wireless sensor network (WSN) [17] consists of a collection of sensors
or nodes capable of monitoring the environment using its local sensors
and by wirelessly communicating with other nodes and to a base sta-
tion. WSNs may vary widely in their topology from simple star or ring
network to complicated multi-hop networks. Each node is designed to
work autonomously using its own battery power. Due to limited battery
power, the nodes are constrained in terms of sensing capability, compu-
tational power and transmission ability. Their major task is to monitor
an environment for a long period of time and hence conserving battery
power by turning off the transmission channels is one of the crucial tech-
niques that need to be used for algorithm development and deployment
in such networks. Over the last decade, the sensor nodes have evolved
a lot in terms of their size and sensing/transmission capability. As a
result, there has been a renewed interest in using sensor networks for a
plethora of applications - forest fire detection, air pollution monitoring,
oceanographic applications, system health monitoring, greenhouse mon-
itoring, battlefield and other military applications to name a few. The
main characteristics of a WSN are:
Limited computation and transmission ability
Frequent and recurrent node failures
Unreliable communication links
Heterogeneity of nodes
Scalability to large scale of deployment
Ability to withstand harsh environmental conditions
Given these constraints, it is easy to see than the standard data min-
ing/machine learning algorithms are not directly applicable to a WSN
setting. As a result, researchers have proposed several algorithms for
modern sensor networks which take into account some or all of these
constraints. One of the main items to consider for WSN is reduce the
sensor communication requirements for broadcasting all the data to the
base station and, it is in this context, that distributed data mining is
likely to play a major role. The major goal of such distributed algo-
rithms is to develop methods so that a node first does some local com-
putation on its own data, and then communicates with nearby neighbors
(in-network processing) to compute a global model. In this chapter, we
present a sampling of three important topics of distributed data mining
