Understanding Mobile Network Failures through Tensor Factorizations

The thesis will focus on the analysis of various network metrics (such as outages, packet loss, etc.) and address the problem of identification of spatial and temporal patterns in the data in order to understand the underlying causes of outages and performance degradations in mobile networks.
Master

Vast amounts of data is being collected to assess the reliability and performance of mobile broadband network and needs to be processed in order to understand the potential causes of failures [1, 2]. We started collecting this data in 2013 and it amounts to over 13 TB in volume. Analysis of such data is a challenging task since these data sets often contain missing data and have different time granularity, in addition to having multiple axes of variation such as various metrics being collected in time from multiple nodes. Such data sets can be represented as multi-way arrays (also referred to as tensors), and tensor factorizations have proved useful in terms of capturing the underlying patterns in such higher-order data sets in different disciplines including signal processing, neuroscience, and social networks [3].

Goal

The goal of the project is to identify spatial and temporal failure patterns in the data using the right data representation and data mining technique. We plan to study various representations of the data and assess the performance of various factorization methods in terms of revealing the failure patterns. In
particular, we aim to identify spatial correlation in outages within a particular network operator like Telenor or between operators, for example
between Telenor and Telia. We would like also to quantify the impact of routine maintenance activities on network reliability and user experience. The identified patterns will be validated using outage data provided by the network operators.

This project will contribute to improving our understanding of complex failure patterns. The gained insights will be communicated to the involved network
operators and other relevant stakeholders like the national telecom regulator (NK

Learning outcome

The thesis will develop skills in two different areas: (i) network communications: working and understanding network data, (ii) data mining: basic data analysis skills as well as understanding of matrix and tensor factorizations, and their use in an application.

Qualifications

Linear algebra and programming skills (BS in Computer Engineering, excellent oral and written English skills).

Supervisors

  • Ahmed Elmokashfi
  • Evrim Acar
  • Olav Lysne

Collaboration partners

Telia, Ice.net

References

  1. robustenett.no/map
  2. A. Elmokashfi, D. Zhou, D. Baltrünas. Adding the next nine: An investigation of mobile broadband networks availability. Proceedings of the 23rd ACM MobiCom, pp. 88-100, 2017.
  3. E. Acar, B. Yener. Unsupervised Multiway Data Analysis: A Literature Survey, IEEE Transactions on Knowledge and Data Engineering, 21(1): 6-20, 2009.

Contact person