Improved Detection of DoS Attacks using Intelligent Computation Techniques

Abstract

IDSs play a principal role in pro-actively detecting intrusions into enterprise-level computer networks, therefore the accuracy with which it performs this vital function is of paramount importance. Many studies have previously been conducted to improve upon proper classification of detections using neural networks and machine learning algorithms. We try to compare the performance of various intelligent computation techniques like Bayesian networks, Naïve Bayesian, Logistic regression, RBF networks, Multi-Layer perception, SVMs with the SMO model, Kth nearest neighbour and Random forest in detecting DoS attack patterns. The data that was used to train and validate these techniques was obtained from the MIT Lincoln lab study into IDSs. The results obtained provide a clear comparison of the individual intelligent computation techniques ability in identifying and classifying attack patterns. Keywords: Networks, intrusion detection, denial of service, datasets, data mining, Bayesian networks, Naïve Bayesian, Logistic regression, RBF networks, Multi-layer perception, Support vector machines, Sequential minimal optimization, Kth nearest neighbor, Random forest.

View PDF