Can a Negative Selection Detect an Extremely few Non-self among Enormous Amount of Self Cells?

We have had lots of reports in which they asserted a negative selection algorithm successfully distinguished non-self cells from self cells, especially in a context of “network intrusion detection” where self patterns are assumed to represent normal transactions while non-self patterns represent anomaly. Furthermore they went on to assert a negative selection gives us an advantage that we use only a set of self cells as training samples. This would be really an advantage since we usually don’t know what do anomaly patterns look like until they complete an intrusion when it’s too late. We, however, suspect, more or less, its applicability to a real system. This paper gives it a consideration to one of the latest such approaches.