Hybrid artificial neural network and gravitational search algorithm in intrusion detection system
Network Intrusion Detection System (IDS) is an automated system that can detect a malicious traffic and nowadays, IDS becomes one of the most important part of computer security systems. But, unfortunately IDS suffers from major problems, which researchers tried to cover them. These problems are low...
Saved in:
Main Author: | |
---|---|
Format: | Thesis |
Published: |
2013
|
Subjects: | |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Network Intrusion Detection System (IDS) is an automated system that can detect a malicious traffic and nowadays, IDS becomes one of the most important part of computer security systems. But, unfortunately IDS suffers from major problems, which researchers tried to cover them. These problems are low detection accuracy and high false alarm rates. There are various approaches being utilized in intrusion detection systems; the first method was using ANN in order to solve complex problem in IDS. But unfortunately, ANN is inefficient in training which may result in poor ANN structure, so, that will give poor detection accuracy. These days, in some cases hybrid NN - Genetic Algorithm and hybrid NN - Particle Swarm Optimization was implemented on IDS in order to qualify IDS to detect intrusions, but unfortunately none of the systems is completely faultless so far, therefore, the search for improvement continues. Recently GSA has been considered, which, is based on the interaction of masses in the universe via Newtonian gravity law, and hybrid GSA-ANN becomes more considerable. The purpose of this study is to improve the efficiency of detection in IDS by applying hybrid Artificial Neural Networks algorithm and Gravitational Search Algorithm in order to reach the high detection accuracy and reduce false alarm rates. The benchmark DARPA KDDCup 1999 IDS dataset was used. Two types of classification are considered in this study, which are binary output classes’ classifier and multiple output classes, classifier. After analyzing the results, the accuracy of GSA-ANN in binary output classes were evaluated (99.73%) for Normal class, (98.4%) for Probe class, (89.95%) for DOS class, (77.97%) for U2R class and (99.34%) for R2L class. In multiple output classes’ classifier the accuracy of GSA-ANN was (95.63%). |
---|