I have a similar problem area as mentioned Trajectory Clustering: Which Clustering Method? . As per suggestion to use dynamic time warping for aligning the time series, are there any differences in performance when a data is clustered with k means and Gaussian mixture model (GMM) or in general when to employ k means and GMM for clustering?Under what circumstances should k means and GMM be chosen for clustering? Also, in the response mentioned how is the centroid calculated from (x,y) ?


Both k-means and GMM need to be able to compute a sensible mean of the cluster, and are designed for Euclidean distance.

This is not really trivial for trajectories; you may need to do a lot of interpolation.

Instead, use a distance based clustering that can handle DTW distances; such as DBSCAN, OPTICS and good old hierarchical linkage clustering.

See also: https://stats.stackexchange.com/questions/70867/how-to-get-the-centroid-of-one-cluster


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.