Using Support Vector Machines in Anomaly Intrusion Detection
Eric Nyakundi · The Atrium (University of Guelph) · 2015
Recent increase in hacks and computer network attacks around the world, including Sony Pictures (2014), Home Depot (2014), and Target (2014) gives a compelling need to develop better Intrusion Detection and Prevention systems. Network intrusions have become larger and more pervasive in nature. However, most anomaly intrusion detection systems are plagued by large number of false positives thus limiting their use. In this Thesis as a contribution to building better Intrusion Detection Systems, we classify intrusions using Support Vector Machines and perform experiments to determine their performance and compare them to other classifiers e.g naive-Bayes, multilayer perceptrons on the network intrusion detection classification task. The classifiers are evaluated on the ISCX2012 dataset. The proposed Support Vector Machine classifier achieves 99.1% average detection accuracy which demonstrates better performance compared to the modified gravitational search algorithm (MGSA) neural network which achieved 97.8% accuracy and the multi-objective genetic algorithm (MOGA) multilayer perceptron which achieved 97% average detection accuracy.