Multivariate Time Series

This is a minimal working example that demonstrates how multivariate time series can be clustered with the pdc package. Note that pdc can only capture limited information in multivariate time series and there are likely other approaches that are better suited to deal with multivariate time series.

First, we load the package and simulate six two-dimensional time series, each with 5000 observations. The first pair are simulated from an autoregressive process with AR=0.1, the second pair are simulated with AR=0.4.

Note that to this end, data must be arranged in a three-dimensional array, with the first dimension representing time, the second dimension identity of the time series, and the third dimension the dimension of the time series. Here, we create data according to dim=c(5000,6,2) that is 6 unique time series with each two channels and 5,000 observed time points.

library(pdc)

set.seed(902101)

ar_params <- rep(c(0.1,0.4), each=3)

X <- array(data = NA, dim=c(5000,6,2))
for (i in 1:dim(X)[2]) {
  for (j in 1:dim(X)[3]) {
    X[,i,j] <- arima.sim(list(ar=ar_params[i]),n=dim(X)[1])
  }
}

labels <- paste("2D-ARIMA ",ar_params,sep="")

Run clustering and plot result

result <- pdclust(X, clustering.method = "single")
plot(result,timeseries.as.labels = FALSE, labels = labels)

Andreas M. Brandmaier

2023-02-17