Can an Immuno-fuzzy Approach Detect Only a Few Non-self Cells Existed in an Enormous Amount of Self Cells?

In the context of network intrusion detection, we test a lately reported technique which generates a set of fuzzy rules to recognize unknown abnormal patterns using a test- function which we call a-tiny-island-in-a-huge-lake. Our concern is whether or not we can train the system only with a set of already known normal patterns. Yet another of our concern is what happens in an extreme case where a sample of abnormal patterns are very few comparing to the normal ones, and what if it eventually shrinks to zero, which is what they call a-needle-in-a-haystack.