interpret plot of multivariate time series clustering results from dtwclust

goyangiee 0
Viewed 34

I'm using DTWCLUST package in r for multivariate time series clustering. Here's my code.

data("uciCT")

mvc <- tsclust(CharTrajMV, k = 4L, distance = "gak", seed = 390L)
plot(mvc)

The CharTrajMV data set has 100 observations with 3 variables. As I understand, clusters are determined based on 3 variables as opposed to univariate time series clustering. Each cluster graph shows several similarly patterned time series (observations) belonging to that cluster. How is this graph drawn? There are 3 time series variables used for clustering, how does one pattern graph come out? I mean the input is 3-dimentional(variables) dataset, but the output is 1-dimentional.

Moreover, I can get the 3 variables's centroid for each cluster (using mvc@centroids)

plot(mvc, labels = list(nudge_x = -10, nudge_y = 1), type="centroids")

this code shows only one centroid for each cluster. Can I get 3 variables' centroid graphs for each cluster with plot option? or is this right approach?

multivariate-time-series cluster-analysis r
1 Answers

This is covered in the documentation. Plotting so many different series in separate panes would get very congested, so, for multivariate plots, the variables are appended one after the other, and you get vertical dotted lines to see the place where that happened, maybe injecting some missing values in some places to account for differences in length. This does mean the x axis isn't so meaningful anymore, but it's only meant to be a quick visualization aid.

userHhjo0
Related