Network attack detection based on combination of neural, immune and neuro-fuzzy classifiers

The paper considers an approach for detection of anomalous patterns of network connections using artificial neural networks, immune systems, neuro-fuzzy classifiers and their combination. The principal component analysis is proposed to optimize the assigned problem. The architecture of the intrusion detection system, based on the application of the proposed methods, is described. The main advantage of the developed approach to intrusion detection is a multi-level analysis technique: first, signature based analysis is carried out, then a combination of adaptive detectors is involved. A number of computational experiments is performed. These experiments demonstrate the effectiveness of the chosen methods in terms of false positive, true positive and correct classification rates.