JOURNAL OF COMPUTERS (JCP)
ISSN : 1796-203X
Volume : 4    Issue : 5    Date : May 2009

On Classification Approaches for Misbehavior Detection in Wireless Sensor Networks
Matthias Becker, Martin Drozda, Sven Schaust, Sebastian Bohlmann, and Helena Szczerbicka
Page(s): 357-365
Full Text:
PDF (450 KB)


Abstract
Adding security mechanisms to computer and communication systems without degrading their
performance is a difficult task. This holds especially for wireless sensor networks, which due to
their design are especially vulnerable to intrusion or attack. It is therefore important to find security
mechanisms which deal with the limited resources of such systems in terms of energy
consumption, computational capabilities and memory requirements. In this document we discuss
and evaluate several learning algorithms according to their suitability for intrusion and attack
detection. Learning algorithms subject to evaluation include bio-inspired approaches such as
Artificial Immune Systems or Neural Networks, and classical such as Decision Trees, Bayes
classifier, Support Vector Machines, k-Nearest Neighbors and others. We conclude that, in our
setup, the more simplistic approaches such as Decision Trees or Bayes classifier offer a
reasonable performance. The performance was, however, found to be significantly dependent on
the feature representation.

Index Terms
Attack, Intrusion and Anomaly Detection; Wireless Sensor Networks; Artificial Immune Systems;
Machine Learning; Bio-Inspired Approach.