Principal Component Analysis of Mobility Data from an Operational GPRS Network

We present a preliminary analysis of mobility data collected from an operational GPRS network. The input data are time-series counting the number of mobile stations present in each of 126 sample routing areas at equally spaced instants (5 min) during one full week. The time-series were extracted from packet-level traces captured by passively monitoring a subset of the Gb links of the network of Mobilkom Austria AG & Co KG during October 2004. We apply the principal component analysis (PCA) to this dataset. The PCA offers a simple method for classifying the routing areas into two main groups, residential and business areas, plus a few "atypical" ones. Additionally, we address the problem of robustness of the PCA to temporary local gaps in the input data