TSclust: AnRPackage for Time Series Clustering

Abstract

Time series clustering is an active research area with applications in a wide range of fields. One key component in cluster analysis is determining a proper dissimilarity measure between two data objects, and many criteria have been proposed in the literature to assess dissimilarity between two time series. The R package TSclust is aimed to implement a large set of well-established peer-reviewed time series dissimilarity measures, including measures based on raw data, extracted features, underlying parametric models, complexity levels, and forecast behaviors. Computation of these measures allows the user to perform clustering by using conventional clustering algorithms. TSclust also includes a clustering procedure based on p values from checking the equality of generating models, and some utilities to evaluate cluster solutions. The implemented dissimilarity functions are accessible individually for an easier extension and possible use out of the clustering context. The main features of TSclust are described and examples of its use are presented.

Data-driven demand response characterization and quantification

“…The color codes the partition into clusters. An increase of standard deviation with average d.Cor results from the non-linearity of d.Cor [10]. The standard deviation is mostly used to discriminate between houses with the same average participation but different variability over the heating season.…”

“…Using the correlation-based distance (d.Cor) in (2) between the individual loads x n and the RTP p of a single day, the response during that day can be quantified [10]. Note that an average correlation of 0 over a day corresponds to a d.Cor of one and the more negative the correlation is, the larger the d.Cor.…”

Monthly Rainfall-Runoff Modeling at Watershed Scale: A Comparative Study of Data-Driven and Theory-Driven Approaches

“…To identify common patterns in these various complicated responses, we conducted clustering analysis on the individual sensitivity curves. We used hierarchical clustering [58] with the correlation distance [59] as the measure of similarity. The correlation distance, given as 2(1 − ρ), where ρ is the correlation between two curves being compared, is an increasing function of the correlation coefficient, hence clustering based on this distance metric groups curves with similar shapes into the same cluster.…”

Analysis of the Temporal Structure Evolution of Physical Systems with the Self-Organising Tree Algorithm (SOTA): Application for Validating Neural Network Systems on Adaptive Optics Data before On-Sky Implementation

“…Usually, clustering comparison is performed in a one to one contrast, giving 0 or 1 if two series are in different or same cluster respectively, and averaging [31,32]. When the distance between clusters increases, the structure presents more differences, so we want to take this into account.…”