DBSCAN for trajectories about dbscan

[mhahsler]

If all your trajectories have the same length then you would just create a matrix where each row has all the x and y values of the points defining the trajectory.

[kageazusa]

Thank you! Let me ask you additional questions:

1. Is it the same to treat the data as {(x1, y1), (x2, y2) ..., (xn, yn)} and as (x1, x2, ..., xn, y1, y2, ..., yn) in the context of DBSCAN? Your answer is the latter case, but I'm not sure if they are identical.
2. What if I would like to analyze trajectories with different lengths? Does a matrix with NA work? (e.g. trajectory 1: length 5, trajectory 2: length 3 -> Can I fill the tail of the row of trajectory 2 with NA?)

[mhahsler]

Different lengths are an issue and NAs will not work. The best solution would be to calculate a distance matrix between the trajectories using the appropriate approach (see e.g., https://zheng-kai.com/paper/vldbj_2019.pdf) and then do dbscan on that distance matrix.

[kageazusa]

Thank you very much. I will look over the paper!