The Hierarchical Spectral Merger Algorithm: A New Time Series Clustering Procedure
KAUST DepartmentComputer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division
Online Publication Date2018-04-12
Print Publication Date2018-04
Permanent link to this recordhttp://hdl.handle.net/10754/627481
MetadataShow full item record
AbstractWe present a new method for time series clustering which we call the Hierarchical Spectral Merger (HSM) method. This procedure is based on the spectral theory of time series and identifies series that share similar oscillations or waveforms. The extent of similarity between a pair of time series is measured using the total variation distance between their estimated spectral densities. At each step of the algorithm, every time two clusters merge, a new spectral density is estimated using the whole information present in both clusters, which is representative of all the series in the new cluster. The method is implemented in an R package HSMClust. We present two applications of the HSM method, one to data coming from wave-height measurements in oceanography and the other to electroencefalogram (EEG) data.
CitationEuán C, Ombao H, Ortega J (2018) The Hierarchical Spectral Merger Algorithm: A New Time Series Clustering Procedure. Journal of Classification. Available: http://dx.doi.org/10.1007/s00357-018-9250-5.
SponsorsThis work was partially supported by 1) CONACYT, México, scholarship as visiting research student, 2) CONACYT, México, Proyectos 169175 Análisis Estadístico de Olas Marinas, Fase II and 234057 Análisis Espectral, Datos Funcionales y Aplicaciones, and 3) Centro de Investigación en Matemáticas (CIMAT), A.C. Euán and Ortega wish to thank Prof Pedro C. Alvarez Esteban for several fruitful conversations on the methodology proposed on this paper. C. Euán wishes to thank to UC Irvine Space Time Modeling Group for the invitation to collaborate as a visiting scholar in their research group. The research conducted at the UC Irvine Space-Time Modeling group (PI: Ombao) is supported in part by the National Science Foundation Division of Mathematical Sciences and the Division of Social and Economic Sciences. The authors thank Dr. Steven C. Cramer of the UC Irvine Department of Neurology for sharing the EEG data that was used in this paper. This work was done while J.O. was visiting, on sabbatical leave from CIMAT and with support from CONACYT, México, the Departamento de Estadística e I.O., Universidad de Valladolid. Their hospitality and support is gratefully acknowledged.
JournalJournal of Classification