Characterisation of nuisance calls in-network and improved detection using machine learning techniques

Nuisance calls are a significant global problem. In this project the student will be working alongside British Telecommunications (BT) to further the understanding of the in-network characteristics of nuisance calls extending our understanding of how advanced statistical and machine learning techniques can be used to better discriminate between classes of nuisance calls and legitimate traffic over the BT network.

The student will have access to a big data environment and observations from the BT voice network to generate insights into the nature of nuisance calls. Previous work at BT has focussed on time series analytics using similarity searches based on Dynamic Time Warping (DTW) and Symbolic Aggregate approXimation (SAX) to look for suspected nuisance calls in time series data generated from in-network parameters.

This PhD project will consist of two stages: firstly, identifying features associated with nuisance calls and then, in the second stage, using these features to identify nuisance calls over the network. In contrast to the problem of detection of spam over email, detecting spam over the voice network is still relatively unexplored.

This project will investigate the characteristics of nuisance calls over the voice network using methods such as unsupervised machine learning and graph theory to examine the relationships in-network. The use of geometrical representations and visualisations will be explored to reveal the underlying nature of nuisance calls furthering quantitative and qualitative understanding of the problem.