Machine Learning Approach for Sequence Clustering with Applications to Ground-Motion Selection

Abstract

Clustering analysis of sequential data is of great interest and importance in many science and engineering areas thanks to the explosive growth of time-series data. Effective methods, especially for sequence clustering, are strongly needed to extract features from data for better representation learning. This paper presents an unsupervised machine learning algorithm for sequence clustering based on dynamic k-means. Specifically, the clustering problem is firstly formulated rigorously to an optimization problem, which is then solved by a proposed three-step alternating-direction optimization approach. The performance of the proposed approach is successfully illustrated through three examples with both synthetic data sets and field ground-motion measurements. In particular, this approach is applied to ground-motion clustering/selection and shows satisfactory results. Overall, the results demonstrate that the proposed algorithm is able to effectively cluster sequential data through mining latent inherent characteristics.