Novelty detection in a changing environment: A negative selection approach

In the recent past, there have been a number of engineering studies motivated by analogies with the human immune system. The immune system has provided a rich source of inspiration for pattern recognition, machine learning and data mining analyses. One of the properties of the immune system which proves particularly useful for novelty detection is that of self/non-self discrimination and this forms the basis of the negative selection algorithm which has previously been applied by other researchers to the problem of time-series novelty detection. The object of the current paper is to apply the negative selection algorithm to more general feature sets and also to consider the case of novelty detection where the normal condition set is significantly non-Gaussian or varies with operational or environmental conditions.