Using an Inductive Learning Algorithm to Improve Antibody Generation in a Single Packet Computer Defense Immune System

摘要

The United States Air Force relies heavily on computer networks for many day-to-day activities. Many of these networks are affected by various types of attacks that can be launched from anywhere on the globe. The rising prominence of organizations such as the AFCERT and the MAJCOM NOSCs is evidence of an increasing realization among the Air Force leadership that protecting our computer networks is vitally important. A critical requirement for protecting our networks is the ability to detect attacks and intrusion attempts. This research is an effort to refine a portion of an AFIT-developed intrusion detection system known as the COmputer Defense Immune System (CDIS). CDIS is based on the human immune system and uses antibodies to attempt to detect network intrusion attempts. The antibodies have various attributes of which a subset is randomly activated at generation time. This research attempts to determine which of the antibody attributes are most useful in helping to build successful antibodies.