Title: Lagged exact Bayesian online change point detection with parameter estimation
Authors: Michael Byrd - Southern Methodist University (United States)
Linh Nghiem - Australian National University (Australia)
Jing Cao - Southern Methodist University (United States) [presenting]
Abstract: Identifying changes in the generative process of sequential data, known as change point detection, has become an increasingly important topic for a wide variety of fields. A recently developed approach, which we call EXact Online Bayesian Change-point Detection (EXO), has shown reasonable results with efficient computation for real time updates. The method is based on a forward recursive message-passing algorithm. However, the detected change points from these methods are unstable. We propose a new algorithm called Lagged EXact Online Bayesian Change point Detection (LEXO) that improves the accuracy and stability of the detection by incorporating l-time lags to the inference. The new algorithm adds a recursive backward step to the forward EXO and has computational complexity linear in the number of added lags. Parameter estimation is also addressed. Simulation studies with three common change point models show that the detected change points from LEXO are much more stable, and parameter estimates from LEXO have considerably lower MSE than EXO. We illustrate the applicability of the methods with two real world data examples comparing the EXO and LEXO.