Researchers use machine learning to detect fraudulent behaviour in cellular networks
September 22, 2016
California researchers are using machine learning to detect anomalies in real time in streaming cellular network data to help prevent fraudulent use. The work is being presented in a joint paper by Argyle Data and Carnegie Mellon University Silicon Valley’s Department of Electrical & Computer Engineering.
Fraudulent use of cellular networks is a growing threat for both network subscribers and operators that costs the industry an estimated $38bn a year. Other emerging consumer behaviour including over-the-top applications present a growing challenge to operator revenues. This has created an increasingly urgent need for robust analytics and detection capable of identifying anomalous behaviour and adapting to evolving network usage patterns in real time.
The research paper “Real-time Anomaly Detection in Streaming Cellular Network Data” will be submitted for presentation at academic conferences during the first half of 2017.
“The sub-field of machine learning known as anomaly detection offers many attractive attributes for providing such solutions,” said the paper’s senior author Ole Mengshoel, associate research professor in the electrical and computer engineering department.
Approaches currently used by mobile communications providers to detect fraud typically rely on static rules with pre-set thresholds. Moreover, such methods cannot address issues on the data plane. However, since more and more fraud will occur on the data network in future, gaining visibility into the characteristics of data usage will be paramount. Due to the vast amount of data flowing across telecoms networks, big data analytics capabilities and the ability to analyse these using advanced machine learning are essential.
“This approach described in this paper is unique,” said Padraig Stapleton, VP of engineering at Argyle Data. “It describes a totally new machine learning method that includes significant developments to create a lightweight product that is fast to train and offers state-of-the-art accuracy as well as other features to help analysts make rapid decisions, all of which are essential for operators’ production environments.”
In this work, Mengshoel and first author David Staub, data scientist at Argyle Data, propose and validate a supervised and unsupervised machine learning-based approach that automatically learns the difference between normal and anomalous call patterns based on usage data.
Co-authors of the paper are Aniruddha Basak, PhD candidate, CMU, Wendy Fong, senior strategic programmes manager, CMU, and machine learning expert Arshak Navruzyan.