Date of Award
Master of Science
Theses & dissertations (College of Business)
Intrusion detection systems help computer systems prepare for and deal with malicious attacks. They collect information from a variety of systems and network sources, then analyze the information for signs of intrusion and misuse. A variety of techniques have been employed to analyze the information from traditional statistical methods to new emerged data mining approaches. In this thesis, we describe several algorithms designed for this task, including neural networks, rule induction with C4.5, and Rough sets methods. We compare the classification accuracy of the various methods in a set of UNIX process execution traces. We used two kinds of evaluation methods. The first evaluation criterion characterizes performances over a set of individual classifications in terms of average testing accuracy rate. The second measures the true and false positive rates of the classification output over certain threshold. Experiments were run on data sets of system calls created by synthetic sendmail programs. There were two types of representation methods used. Different combinations of parameters were tested during the experiment. Results indicate that for a wide range of conditions, Rough sets have higher classification accuracy than that of Neural networks and C4.5. In terms of true and false positive evaluations, Rough sets and Neural networks turned out to be better than C4.5.
Zhang, Xiaoning, "Data mining approaches for detecting intrusion using UNIX process execution traces" (2000). Retrospective Theses and Dissertations. 17699.