A0903
Title: Detecting change-points in noisy data sequences with continuous piecewise structures
Authors: Yiming Ma - University of Otago (New Zealand) [presenting]
Andreas Anastasiou - University of Cyprus (Cyprus)
Ting Wang - University of Otago (New Zealand)
Fabien Montiel - University of Otago (New Zealand)
Abstract: A new method, called singular spectrum analysis isolate-detect (SSAID), is proposed to detect change-points in noisy data sequences with an underlying continuous piecewise structure. In contrast to existing parametric change-point detection methods for signals with a predefined piecewise structure, SSAID does not require prior knowledge of the exact nature of the structural changes; for example, it can identify change-points in noisy piecewise-exponential or piecewise-quadratic signals equally well. SSAID is motivated by the need for automated detection of slow slip events (SSEs), which are a type of slow earthquakes. The SSE data have a typical piecewise-non-linear trend, but the exact structure is unknown. SSAID recasts the problem of identifying SSEs as that of detecting change-points in a piecewise-linear signal. This is achieved by obscuring the deviation from the piecewise-linear in the underlying piecewise non-linear signal. The results on both simulated and real SSE data, and simulated data with various piecewise structures suggest that our method can successfully detect change-points in signals with a wide range of piecewise structures. We further demonstrate the performance of SSAID using real data such as the number of COVID-19 daily confirmed cases in the United States and the monthly S\&P 500 close price index.
