EcoSta 2017: Start Registration
View Submission - EcoSta2017
A0384
Title: An efficient analysis of change points via swarm intelligence Authors:  Hsin-Hao Chen - Academia Sinica (Taiwan)
Frederick Kin Hing Phoa - Academia Sinica (Taiwan) [presenting]
Livia Lin-Hsuan Chang - Academia Sinica (Taiwan)
Abstract: Evolutionary algorithm is a new and promising method to statistical optimization. Recently, a nature-inspired metaheuristic method was proposed, namely the Swarm Intelligence Based (SIB) method, for efficient optimization in discrete and continuous domains, but it is of restricted use due to fixed particle size. We introduce a new operation called VARY to the standard SIB framework. It allows the adjustment on the number of change points to be included during the optimization process. We apply the enhanced algorithm for analyzing composite functions that consist of multiple change points. Numerical results show that our algorithm accurately detect the location of change points by using a small number of particles and iterations. Comparing to existing methods in the literature, our algorithm outperforms them in terms of accurate number and location of change points. Our method is demonstrated on the analysis of the data of global surface temperature, indicating that 1906, 1945 and 1963 are the three change-point years for the global surface temperature.