An Adaptive DCT Based Intrusion Detection System

Anomaly based intrusion detection is a critical research area, since it does not require any prior knowledge of attack signatures in order to detect them. Typically, intrusion detection systems (IDSs) consider a fix learning model, and generally do not contemplate any memory constraint. These assumptions may not be valid in real world scenarios, where the learning model may evolve and the available memory is restricted. In this paper a cluster based method has been proposed which uses Discrete Cosine Transform (DCT) to build an effective and compact model of normal data. This model is then used for intrusion detection in environments where there is a concept drift. The proposed method has been evaluated with the KDD99 intrusion detection dataset. Simulation results indicate the superiority of the proposed method when compared to a state-of-the-art IDS. Keywords-component; IDS, Novelty detection, DCT.