Mobile traffic dataset comparisons throughcluster analysis of radio network event sequences

Abstract: Ericsson regularly collects traffic datasets from different radio networks around the world. These data sets can be used for several research purposes, ranging from general statistics to more specific studies such as system troubleshooting and buffer-level analysis. Currently, a researcher may find it difficult to assess if a certain dataset is useful for aparticular investigation, since there exists no easily accessible overview of the properties of the different data sets.This thesis project aims to make it easier to compare the existing traffic datasets in terms of general statistics, user and time coverage,data integrity and the patterns of sequences in radio network event logs. The key contribution is a method of clustering event sequences based on sequence duration and occurrences of a number of key events.A method called the Gap-statistic was applied to determine that using 11 clusters was suitable for the analysis, although no strong evidence was found for the existence of well separated clusters.The results show that the method can work as a useful extension of basic comparative statistics. Two dense ranges of sequence durations discovered in the basic statistics could successfully be linked to corresponding clusters of sequences. Extensive statistics about the cluster members then revealed detailed properties of the sequences in these two dense areas, at a deeper level than could be understood from the basic statistics.A problematic part of interpreting the results of the method is that many different perspectives of the data need to be considered at the same time to find interesting links. Future work could include automating the process of linking features in the basic statistics to clusters.