A Case Study of ICA with Multi-scale PCA of Simulated Traffic Data

Often packet traffic data is non-stationary and non-gaussian. These data complexity causes difficulties in its analysis by standard techniques and new methods must be employed. Recent theoretical and applied works have demonstrated the appropriateness of wavelets for analyzing multivariate signals containing non-stationarity and non-gaussianity. This paper presents a new pre-processing method, a multi-scale PCA that combines a wavelet filtering method with principal component analysis (PCA), for a noise free independent component analysis (ICA) model. By applying the proposed method to a set of test data coining from simulations of a packet switching network (PSN) model we see improvements of data analysis results.