查看文章

We present a contribution to the network intrusion detection process using Adaptive Resonance Theory (ART1), a type of Artificial Neural Networks (ANN) with binary input unsupervised training. In this phase, we present a feature selection using data mining techniques, towards two dimensional dataset reduction that is efficient for the initial and on-going training. The well know KDD'99 Intrusion Detection Dataset (KDD'99 dataset for short) is tremendously huge and has been reported by many researchers to have unjustified redundancy, this makes adaptive learning process very time consuming and possibly infeasible. We intend to reduce the dataset both vertically and horizontally, numbers of vectors and number of features, such that nearly 10% of the training subset is used for the initial unsupervised training process and nearly 1% of the training subset is used for the on-going training, and that is only regarding the number of vectors. On top of that, only the significant features will be used yielding a highly reduced dataset, and this is the scope of this work.