Application of feast (Feature Selection Toolbox) in ids (Intrusion detection Systems)

Abstract

Security in computer networks has become a critical point for many organizations, but keeping data integrity demands time and large economic investments, in consequence there has been several solution approaches between hardware and software but sometimes these has become inefficient for attacks detection. This paper presents research results obtained implementing algorithms from FEAST, a Matlab Toolbox with the purpose of selecting the method with better precision results for different attacks detection using the least number of features. The Data Set NSL-KDD was taken as reference. The Relief method obtained the best precision levels for attack detection: 86.20%(NORMAL), 85.71% (DOS), 88.42% (PROBE), 93.11%(U2R), 90.07(R2L), which makes it a promising technique for features selection in data network intrusions.