Intrusion Detection Based on Feature Transform Using Neural Network

In this paper, a novel method for intrusion detection is presented. The presented method uses a clustering method based on the transformed features, which can enhance the effectiveness of the clustering. Clustering is used in anomaly detection systems to separate attack and normal samples. In general, separating attack and normal samples in the original input space is not an easy task. For better separation of the samples, a transformation that maps input data into a different feature space should be performed. In this paper, we propose a novel method for obtaining proper transformation function that reflects the characteristics of the given domain. The transformation function is obtained from the hidden layer of the trained three-layer neural network. Experiments over network connection records from KDD CUP 1999 data set are used to evaluate the proposed method. The result of the experiment clearly shows the outstanding performance of the proposed method.