Intrusion Detection System for Self Configurable Networks

Abstract

This thesis makes four contributions which are described following in order. newline Module 1 is designed to obtain the IDS system based on the bioinspired immune system. The aim of this module is to create a framework that can protect PC systems from infections and other forms of digital attack. The proposed frameworks efficiency is measured in terms of protecting computers from unapproved intruders. newlineModule 2 is designed for an efficient IDS using supervised machine learning. The used machine learning approaches: SVM and kNN are responsible for distinguishing normal data packets from attack data packets using the principal component analysis methodology. Thus the proposed approach is termed PCA kNN. newlineModule 3 is designed for an efficient IDS using SVM. The designed machine learning-based model is tested with kernel functions such as Linear Sigmoidal and Redial functions to improve the accuracy in the results. The proposed solution decreases the effect of a Denial of Service attack, which leads to a Denial of Sleep attack, according to observations. Positive predictive value true positive rate and overall classification accuracy are often used to assess the success of the proposed process. newlineModule 4 proposed a system that utilize SVM, probabilistic neural network Decision Tree NFC smooth SVM and kNN classifiers in a hierarchical manner. To create a hybrid hierarchical classifier, the best performing classifiers at each level are ordered in hierarchical order. Thus, the overall detection ratio is high at each level. newline The thesis overall goal is to improve the security of self configurable networks. Any practical difficulties faced by network administrators are taken into consideration when formulating the objectives. Machine learning based intrusion detection systems are designed and implemented in order to meet the thesis objectives newline newline