| With the growth of the Internet in regards to the exchange of information, security has become a growing concern. Hackers constantly find new variants of types of intrusions to gain access to systems. In order to prevent these new attacks from succeeding, predictive algorithms that can classify based on probabilistic models are needed. Recent gains have been made using data mining algorithms to recognize patterns in large datasets.; This thesis describes the benefits of combining classification-association rule data mining with the predictive strengths of a radial-basis function neural network. This new algorithm provides improved predictability of the various categories of intrusion detection beyond what they perform on their own. Additionally, performance is shown to improve on that of a back-propagation neural network. |
