I am new to multivariate time-series and sequences analysis and I cannot find anything relevant to my problem.
I have a dataset with sequential data (each sample of a sequence corresponds to features such as heading, speed, longitude, latitude etc.). The dataset contains many of these sequencies (ship trajectories).
I am trying to do a classification of these trajectories-sequencies using KNN. I have already tried with some distance metrics, but I have been asked to find a way to use correlation between the trajectories combined with the KNN.
What I am thinking is that for trajectory1 (T1) and trajectory2 (T2) I could find the pairwise correlations for the feature sequencies (T1_heading_sequence with T2_heading_sequence, T1_speed_sequence with T2_speed_sequence etc.), calculate the correlation distances between the vectors as 1-r and add them up to find the final dissimilarity measure to use for the KNN classification.
Is this any reasonable? Or if not, could you suggest another dissimilarity measure based on correlation that I could probably use for my problem, so that I can study about it?
Thank you in advance