A Polyscale Autonomous Sliding Window for Cognitive Machine Classification of Malicious Internet Traffic

Features of an Internet traffic time series can be estimated using dynamical systems. Dynamical systems may exhibit chaos and strange attractors [1] [2]. Since Internet traffic shows non stationarity and long term dependence among data samples, a cognitive polyscale approach should be taken to analyze the hidden features in a nonlinear data time series. It is necessary to estimate a reasonable window of time series so that the polyscale analysis can be performed without violating the statistical bounds of the analysis. In this work, a feature extraction algorithm is developed using variance fractal dimension trajectory and the statistical parameters of the calculation are validated using an autonomous varying window of data samples. Our analysis shows promising results since the algorithm is able to capture the presence of DNS denial of service attack and has extracted the bursts of data sample accurately. Keywords— Cognitive machine learning, Fractal, Polyscale, DNS DDoS amplification attacks, Anomaly detection, Cyber threats, Variance fractal dimension, Non stationary trend