Improving the Intrusion Detection Systems' Performance by Correlation as a Sample Selection Method

Due to a growing number of the computer networks in recent years, there has been an increasing interest in the intrusion detection systems (IDSs). In this paper we have proposed a method applied to the instance selection from KDD CUP 99 dataset which is used for evaluating the anomaly detection techniques. In order to determine the performance of proposed method in the dataset reduction, a feed forward neural network was trained by a reduced dataset to classify normal or attack records in the dataset. The most obvious finding resulted from this study is a considerable increase in the accuracy rate obtained from the neural network.