Wavelet transforms and change-point detection algorithms for tracking network traffic fractality

Network traffic is known to present fractal characteristics such as LRD, which can be efficiently measured with wavelet transforms. Current methods of estimation of fractal parameters, such as the LogScale diagram, are not able to track the changes of the parameters. This paper proposes and studies the combined use of wavelet transforms (DWT, MODWT, DTWT) and change-point detection algorithms (ICSS, SIC) in order to detect the instants when fractality changes noticeably. The different approaches are contrasted and statistical assessment is provided, together with the results of the procedure when applied to synthetic traces.