ART Neural Network Clustering Algorithm And Its Application To Intrusion Detection

Intrusion detection system (IDS) is an active protection system for network security. It is a new type of security protection technology following traditional security protection methods such as data encrypting, firewall and so on. IDS offers real-time protection against interior/exterior attacks and mistaken operation, as it can hold up and respond to an intrusion before the host or the network is endangered. However, with the diversity of intrusion techniques, traditional intrusion detection technologies cannot meet the needs of current network environment. Applying artificial intelligent technologies to IDS has been a focus among recent researches.Classification and clustering are two important functions of Data Mining (Knowledge Discovery from Databases). The main difference between them is that the former has a supervised learning mechanism, while the latter has an unsupervised one.After introducing the traditional methods of classification and clustering, this paper gives a particular analysis of the ART-based clustering method. Adaptive Resonance Theory (ART) neural network model is used in data clustering. During its learning process, the ART network can solve the stability/plasticity dilemma well. This feature of ART makes it appropriate as a model to be used in data clustering.By analyzing the standard ART model and intrusion detection respectively, the dissertation presents two modified ART models, in order to make the clustering results orderly and hierarchical when applying ART to intrusion detection. The simulation results based on the KDD Cup99 dataset show that both of the modified models achieve the expectant goal.