A combinatorial optimization algorithm used to find maximum-weight matchings, which can be used to solve the assignment problem.
Published in Chapter:
Segmented Dynamic Time Warping: A Comparative and Applicational Study
Ruizhe Ma (Georgia State University, USA), Azim Ahmadzadeh (Georgia State University, USA), Soukaina Filali Boubrahimi (Georgia State University, USA), and Rafal A. Angryk (Georgia State University, USA)
Copyright: © 2019
|Pages: 19
DOI: 10.4018/978-1-5225-8446-9.ch001
Abstract
Initially used in speech recognition, the dynamic time warping algorithm (DTW) has regained popularity with the widespread use of time series data. While demonstrating good performance, this elastic measure has two significant drawbacks: high computational costs and the possibility of pathological warping paths. Due to the balance between performance and the tightness of restrictions, the effects of many improvement techniques are either limited in effect or use accuracy as a trade-off. In this chapter, the authors discuss segmented-DTW (segDTW) and its applications. The intuition behind significant features is first established. Then considering the variability of different datasets, the relationship between specific global feature selection parameters, feature numbers, and performance are demonstrated. Other than the improvement in computational speed and scalability, another advantage of segDTW is that while it can be a stand-alone heuristic, it can also be easily combined with other DTW improvement methods.