Anomaly Detection in SIGMA data

Ward, Kes and McGarry, Luke and Pyke, Caroline (2023) Anomaly Detection in SIGMA data. In: UNSPECIFIED. (In Press)

Text (Kes__Sigma_Data_Challenge)
Kes_Sigma_Data_Challenge.pdf - Other
Available under License Creative Commons Attribution.
Download (3MB)

Abstract

The Poisson Functional Online Cumulative Sum (Poisson-FOCuS) method is a method for solving the likelihood ratio test of Poisson(λ) null against Poisson(μλ) alternative where μ>1, i.e. searching for an increase in count. This can be thought of as equivalent to testing all possible anomaly start points τ<T at each timestep T, giving a computationally efficient way to analyse count anomalies that occur over intervals of time. We run the Poisson-FOCuS method on SIGMA data, with an additional adjustment to remove anomaly tail traces, and report the results.