At first glance this may seem somewhat innocuous and abstract, but the uses for change detection are incredibly ubiquitous, especially for a company such as BT, where spotting a small shift in a single data series in efficient fashion can make all the difference in mitigating a future operational failure, containing a denial of service attack, or simply maximising speed of customer service.
Changepoint detection has been used elsewhere to great effect, including, health care, finance, environmental science and large-scale retail. The advent of the Big Data Age, however, has presented something of a challenge for existing changepoint methods, where datasets of interest can have millions of variates in addition to a high density of observations.
I am therefore specifically interested in efficient changepoint detection for data streams. This project contains two broad phases
- The first being to examine the means of speeding up current techniques, and
- The second to pioneer novel approaches to multivariate changepoint detection.