I am trying to predict time series. I have several samples which
consist of measurement of 4 different variables over different
time periods, and other measurements over subsequent time
periods which I believe is predicted by the 4 variables.
I am thinking of using cluster analysis, to confirm that
there is a basis for my belief. I therefore need to compare
samples of different lengths, to find the distance between them.
I am asking for suggestions on how to compute a distance
between my samples.
I have thought of using the R time series function spectrum on
each sample, and comparing the spectrum differences using
the ks test function.
I have also thought of using the dtw package to determine a
difference. I could also interpolate so each sample is the same
length, and use shape matching.
I would also like to use either regression or a neural network to
try to predict my measurements. Is anyone aware of prediction
methods which would not require all my samples to be the same
length or how I might deal with different length samples?
Thank you,
Dan
[[alternative HTML version deleted]]