I wish to compare and cluster 10 time-series with different lengths. What are the techniques that I can used and do they have implementation in R?
I found that dynamic time wrapping is nice but does it only work with equal lengths?
Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It only takes a minute to sign up.Sign up to join this community
It just happened that few days ago I read Marco Cuturi's paper on "Fast Global Alignment Kernels" . The idea is to cast the well-known DTW distances as similarities eligible for use in kernel machines, e.g. SVM. You cannot directly transform DTW distance into similarity and hope it will work (e.g. negative exponential of distance) - you will get non positive definite kernel. The author proposed a novel technique with Global Alignment kernels such that the nice DTW properties are accessible as well as the kernel is positive definite.
I have experimented with his code on several standard time-series classification datasets  and was pleasantly surprised of the performance. The beauty and power of his approach that it works for multivariate time-series as well. And oh, it works for different length time-series as well.
Ten time-series might be too little to train a good classifier. Perhaps getting more examples is possible.
The paper, implementations (C, MATLAB and Python) can be found [here].
UPDATE: Author's website has moved. Please use the following link to get paper and implementations .
You can simply extend what I responded in How to statistically compare two time series? to more than two series . Essentially identify a common model and estimate the parameters both globally and locally. Perform an F test and either accept or reject the hypothesis of common parameters. Another post gives a similar view Proving similarities of two time series