Internet Traffic Modeling and Analysis with Application to Cybersecurity: Automated Anomaly Detection, Low Volume Anomaly Detection, Fault IP Address Identification

[] Internet traffic modeling and analysis is critical for network design and for cybersecurity. Internet traffic differs from Telephone traffic insofar as it characterized by long range dependent scale-free temporal dynamics. In this talk, we will describe multiscale analysis as a state-of-the-art tool to assess and quantify scale-free dynamics. We will also that show that wavelet analysis mut be combined with random projection strategies to permit a statistical characterization of Internet background traffic both accurate and robust to anomalies. In turn, these random projections can be further involved into automated anomaly detection and into the identification of the IP addresses involved. However, scale-free analysis remained so far mostly univariate, applied independently to directional counts of either bytes or packets, while challenges in cybersecurity naturally call for multivariate analysis. Elaborating on recent theoretical developments on eigenvalue-based multivariate self-similarity analysis, this talk will provide evidence for multivariate self-similarity in 17 years of Internet traffic data from the MAWI repository and will discuss the potential use of multivariate self-similarity for low volume anomaly detection.
Co-sponsored by: Fairleigh Dickinson University
Speaker(s): Dr. Patrice Abry
Agenda:
Fairleigh Dickinson University
1000 River Road, Building: Muscarelle Center, Room Number: 105
Teaneck, New Jersey, United States 07666
For additional information about the venue and parking, please contact
Dr. Hong Zhao
zhao@fdu.edu
Virtual: https://events.vtools.ieee.org/m/469441