Intrusion detection system based on carpenter grossberg artificial neural network

Other Title(s)

نظام كشف التطفل استنادا على شبكة كاربنتر كروسبيرك العصبونية الاصطناعية

Dissertant

al-Rabii, Ammar Muhanna Kazim

Thesis advisor

Naum, Riyad Shakir

Comitee Members

Kayid, Ahmad
Khalid, Mamun
Hirz Allah, Nail

University

Middle East University

Faculty

Faculty of Information Technology

Department

Computer Science Department

University Country

Jordan

Degree

Master

Degree Date

2014

English Abstract

Over the last few decades, computer applications have evolved and became very important part of our life.

This led to widespread concerns of network service disruption due to large-scale malicious attacks on computer networks.

The development of a secure infrastructure to defend these applications from all challenges coming from intruders, hackers, and unauthorized access is a major challenge.

Intrusion detection system (IDS) is regarded as the second line of defense against network anomalies and threats.

IDSs play an important role in detecting malicious and suspicious activities, and providing warning for unauthorized access over the network.

This research simulates a model of intrusion detection system.

Artificial neural network (ANN) and machine learning (ML) combined with clustering algorithm as a pre-classifier are used to enhance the detection of network intrusion.

This IDS use both Adaptive Resonance Theory (ART1) and k-mean clustering algorithm, where ART1 is a version of Carpenter/Grossberg’s ANN and the key core in this system.

The simulation system includes three main phases: 1.

Preprocessing, in which converts the data and cluster the categories.

2.

Training phase, in which trains ART1 neural network.

3.

Testing phase, which tests ART1 network and check the performance and the stability of the IDS system.

At training phase, the sample space was randomly selected, where all known attack patterns are selected from KDD 99 dataset.

Furthermore, many parameters were adjusted such as; norm, vigilance test, and weight factors.

For testing purposes, the sample space is also randomly selected, that contains a number of duplicated patterns in order to test the stability.

The results of this research has a detection rate about 96.8% with an accuracy rate 96% , false positive rate 1.19% and False Negative Rate about 0.54% The results are compared with other previous studies.

The results from this research showed better performance than the compared approaches.

Main Subjects

Information Technology and Computer Science

No. of Pages

109

Table of Contents

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-699451