Sliding Window Symbolic Regression for Detecting Changes of System Dynamics

• Published in: Genetic Programming Theory and Practice XII pp. 91 - 107
• Main Authors: Winkler, Stephan M., Affenzeller, Michael, Kronberger, Gabriel, Kommenda, Michael, Burlacu, Bogdan, Wagner, Stefan
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing 05.06.2015
• Series: Genetic and Evolutionary Computation
• Subjects: Self-adaptive sliding window techniques; Symbolic regression; System analysis; System dynamics
• ISBN: 331916029X; 9783319160290
• ISSN: 1932-0167
• DOI: 10.1007/978-3-319-16030-6_6

In this chapter we discuss sliding window symbolic regression and its ability to systematically detect changing dynamics in data streams. The sliding window defines the portion of the data visible to the algorithm during training and is moved over the data. The window is moved regularly based on the generations or on the current selection pressure when using offspring selection. The sliding window technique has the effect that population has to adapt to the constantly changing environmental conditions. In the empirical section of this chapter, we focus on detecting change points of analyzed systems’ dynamics. We show its effectiveness on various artificial data sets and discuss the results obtained when the sliding window moved in each generation and when it is moved only when a selection pressure threshold is reached. The results show that sliding window symbolic regression can be used to detect change points in systems dynamics for the considered data sets.