Anomaly detection in noisy time series

The ability to detect anomalies or outliers in time series, and to develop algorithms that can find them at the earliest possible sign of a deviation from expected behaviour, is critical for many applications including those arising in security, epidemiology, and monitoring of critical infrastructure. This project aims to develop new methods to characterise the expected time series behaviour of systems, even in the presence of significant noise and concept drift, and with spatio-temporal and multivariate time series. The goal is to be able to rapidly detect when anomalies are occurring in time for preventive action to be taken.

Project Researchers

Lead CIs

Link to publication

Kandanaarachchi, S. (2017).  Robust anomaly detection in non-stationary data streams. WIMSIG Conference 2017.