Analysis of Network Intrusion Datasets for Attack Classifications
by
Archana Polisetti
University of North Carolina at Greensboro
Coauthors: Shan Suthaharan

Datasets are required for the evaluation of Machine Learning (ML) techniques adopted in the Network Intrusion Detection systems. Several network intrusion datasets are available in the public domain, among them NSL-KDD dataset has been used extensively in the past decade. Although it has been adopted in the development of several ML-based intrusion detection techniques, the studies conducted to learn the classification properties of different intrusions are limited. Therefore we studied the NSL-KDD dataset using statistical variance, runs length technique and Chebyshev inequality, and revealed some interesting classification properties. This research also led to the development of unique 2-bit classifier that classifies the intrusion attacks and normal network traffic robustly.

Date received: August 28, 2012