CMStatistics 2021: Start Registration
View Submission - CMStatistics
B1516
Title: Sequential detection of emergent phenomena within functional data Authors:  Edward Austin - Lancaster University (United Kingdom) [presenting]
Idris Eckley - Lancaster University (United Kingdom)
Lawrence Bardwell - Lancaster (United Kingdom)
Abstract: Detecting anomalies in a sequential setting is a well-studied area of research, however, the sequential detection of anomalies within partially observed functional data, termed emergent anomalies, is an open problem. Classical sequential detection approaches look for changes in the parameters, or structure, of point data and are not equipped to handle the complex nonstationarity and dependency structure of functional data. Existing functional data approaches, on the other hand, require the full observation of the curve before anomaly detection can take place. Motivated by an application arising from telecommunication engineering, we propose a new method that performs sequential detection of anomalies in partially observed functional data. The new method, called FAST, captures the common shape of the curves using Principal Differential Analysis and uses a form of CUSUM test to monitor a new functional observation as it emerges. The performance of FAST is then assessed on both simulated data and telecommunications data, demonstrating the effectiveness of the test in a range of settings.