University of Hertfordshire

By the same authors

Intrusion Detection System using Bayesian Network Modeling

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Documents

View graph of relations
Original languageEnglish
Title of host publication13th European Conference on Information Warfare and Security ECCWS 2014
PublisherACPI (Academic Conference Publishing International)
Pages223-232
ISBN (Print)978-1-910309-24-7
Publication statusPublished - Jul 2014
Event13th European Conf on Cyber Warfare and Security EDDWS 2014 - Piraeus, Greece
Duration: 3 Jul 20144 Jul 2014

Conference

Conference13th European Conf on Cyber Warfare and Security EDDWS 2014
CountryGreece
CityPiraeus
Period3/07/144/07/14

Abstract

Computer Network Security has become a critical and important issue due to ever increasing cyber-crimes. Cybercrimes are spanning from simple piracy crimes to information theft in international terrorism. Defence security agencies and other militarily related organizations are highly concerned about the confidentiality and access control of the stored data. Therefore, it is really important to investigate on Intrusion Detection System (IDS) to detect and prevent cybercrimes to protect these systems. This research proposes a novel distributed IDS to detect and prevent attacks such as denial service, probes, user to root and remote to user attacks. In this work, we propose an IDS based on Bayesian network classification modelling technique. Bayesian networks are popular for adaptive learning, modelling diversity network traffic data for meaningful classification details. The proposed model has an anomaly based IDS with an adaptive learning process. Therefore, Bayesian networks have been applied to build a robust and accurate IDS. The proposed IDS has been evaluated against the KDD DAPRA dataset which was designed for network IDS evaluation. The research methodology consists of four different Bayesian networks as classification models, where each of these classifier models are interconnected and communicated to predict on incoming network traffic data. Each designed Bayesian network model is capable of detecting a major category of attack such as denial of service (DoS). However, all four Bayesian networks work together to pass the information of the classification model to calibrate the IDS system. The proposed IDS shows the ability of detecting novel attacks by continuing learning with different datasets. The testing dataset constructed by sampling the original KDD dataset to contain balance number of attacks and normal connections. The experiments show that the proposed system is effective in detecting attacks in the test dataset and is highly accurate in detecting all major attacks recorded in DARPA dataset. The proposed IDS consists with a promising approach for anomaly based intrusion detection in distributed systems. Furthermore, the practical implementation of the proposed IDS system can be utilized to train and detect attacks in live network traffic

ID: 7351653