Home // ICN 2015, The Fourteenth International Conference on Networks // View article
Authors:
Sandra Kübler
Michael Massoth
Anton Wiens
Torsten Wiens
Keywords: Fraud detection; Voice over IP networks; behavior pattern recognition; unlabeled data; FRITZ!Box.
Abstract:
Widespread monetary losses are known to be caused worldwide by fraud attacks on Voice over IP systems. In 2014, several millions of FRITZ!Box routers have been compromised and used to conduct phone calls to international destinations. By using fraud detection systems, such attacks can be detected. By analyzing Call Detail Records (CDRs), various algorithms can be applied to detect fraud. Unfortunately, this data is mostly unlabeled, meaning no indications on which calls are fraudulent or non-fraudulent exist. In this work, a new method to detect fraud is presented, utilizing the concept of clustering algorithms leading to behavior pattern recognition using information retrieved from user profiles. The grouping aspect of clustering algorithms regarding the similarity of objects leads to data depicting the behavior of a user to be matched against behavior patterns. If a deviation from the assigned behavior patterns occurs, the call is considered fraudulent. A prototype has been implemented with two behavior patterns defined, making it possible to detect fraud. It can further be refined by adjusting multiple thresholds, as well as defining more behavior patterns. The prototype is to be integrated in an existing fraud detection system of Hochschule Darmstadt, being developed in cooperation with a small and medium-sized enterprise (SME) telecommunication provider, improving the quality of its VoIP services.
Pages: 191 to 197
Copyright: Copyright (c) IARIA, 2015
Publication date: April 19, 2015
Published in: conference
ISSN: 2308-4413
ISBN: 978-1-61208-398-8
Location: Barcelona, Spain
Dates: from April 19, 2015 to April 24, 2015