A data mining friendly anonymization scheme for system logs using distance mapping

Abstract

In this document, we investigate the use of Distance Mapping ideas and what they enable for system logs. In the field of telecommunication networks, log files are used for service quality assurance, and are coming from various devices, back end systems, and general usage of the network. Typically, these log files are not allowed to be monetized or shared with third parties, thanks to legal restrictions on privacy issues. While there are some existing early solutions to this, such as Differential Privacy or Homomorphic Encryption, we propose here to look at Distance Mapping to transform the raw data (the system log files) into highly usable but anonymized data. The resulting data can be used directly by Machine Learning algorithms, visualization algorithms, or be considered for re-embedding. While this approach transforms the data format significantly and limits its usage only for distance-based data mining and machine learning tools, it is an elegant and computationally feasible methodology for such applications.