T-detector maturation algorithm with overlap rate
WSEAS Transactions on Computers
Review: The use of computational intelligence in intrusion detection systems: A review
Applied Soft Computing
| Hi-index | 0.00 |
Artificial Immune Systems (AIS) is a research area of developing computational methods inspired by biological immune systems. The approach of negative selection algorithms (NSA) is one of the major models of AIS. This dissertation does a comprehensive survey of NSA and highlights the key components that define a negative selection algorithm. It demonstrates that the so-called 'negative selection algorithms' have been a very broad interpretation compared with its biological archetype and differ from one another in strategy, applicability and implementation. This work proposed a new negative selection algorithm called V-detector. It has several important features that alleviate some difficulties in negative selection algorithms. (1) Statistical techniques are integrated in the detector generation process to estimate the detector coverage. (2) Detectors with variable coverage are used in a highly efficient manner to achieve maximum coverage. (3) A boundary-aware algorithm is proposed to interpret the training data set as a whole, instead of considering them as independent points. It shows that negative selection's certain learning property cannot be replaced by straightforward positive selection. (4) The main components of V-detector can be customized for different data/detector representations and detector generation mechanisms. This generic characteristic could connect the gap between different negative selection algorithms. For example, extension from Euclidean distance to more general distance measures demonstrated its potential to accommodate domain specific elements. (5) One-shot training instead of evolutionary approach is utilized to lead to a more concise model. While it doesn't exclude combination or cooperation with evolutionary process, this simple model makes it possible to implement a very efficient learning process and provides great flexibility for extension. In the light of recent years' doubts about negative selection algorithms, applicability of negative selection algorithms is discussed in details both to understand the reasonable scenarios to use it and its intrinsic limitations. Negative selection algorithms, mainly MILA (Multilevel Immune Learning Algorithm) and V-detector, were experimented on various real-world datasets. To demonstrate its strength, V-detector was used in image-based dental diagnosis with a novel real-valued representation of occlusion condition on dental images.