Metaheuristic-Based Neural Network Training And Feature Selector For Intrusion Detection
Loading...
Date
2019-04
Authors
Mohammed Ghanem, Waheed Ali Hussein
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
Intrusion Detection (ID) in the context of computer networks is an essential
technique in modern defense-in-depth security strategies. As such, Intrusion Detection
Systems (IDSs) have received tremendous attention from security researchers and
professionals. An important concept in ID is anomaly detection, which amounts to the
isolation of normal behavior of network traffic from abnormal (anomaly) events. This
isolation is essentially a classification task, which led researchers to attempt the
application of well-known classifiers from the area of machine learning to intrusion
detection. Neural Networks (NNs) are one of the most popular techniques to perform
non-linear classification, and have been extensively used in the literature to perform
intrusion detection. However, the training datasets usually compose feature sets of
irrelevant or redundant information, which impacts the performance of classification,
and traditional learning algorithms such as backpropagation suffer from known issues,
including slow convergence and the trap of local minimum. Those problems lend
themselves to the realm of optimization. Considering the wide success of swarm
intelligence methods in optimization problems, the main objective of this thesis is to
contribute to the improvement of intrusion detection technology through the
application of swarm-based optimization techniques to the basic problems of selecting
optimal packet features, and optimal training of neural networks on classifying those
features into normal and attack instances. To realize these objectives, the research in
this thesis follows three basic stages, succeeded by extensive evaluations.
Description
Keywords
Neural Network , Intrusion Detection