An Empirical Study of Self/Non-Self Discrimination in Binary Data with a Kernel Estimator

Affinity functions play a major role within the artificial immune system (AIS) framework and crucially bias the performance of AIS algorithms. In the problem domain of self/non-self discrimination by means of negative selection, affinity functions such as the Hamming distance or the r-contiguous distance are frequently applied to measure distances in binary data. In recent years however, several limitations and problems with these distance measurements in negative selection have been identified. We propose to measure distances in binary data by means of probabilities which are modeled with a kernel estimator. Such a probabilistic model is preeminently applicable for the self/non-self discrimination problem. We underpin our proposal with an empirical study on artificially generated and real-world datasets.