Multivariate time series clustering using Dynamic Time Warping (DTW) and k-mediods algorithm This repository contains code for clustering of multivariate time series using DTW and k-mediods algorithm. It contains code for optional use of LB_Keogh method for large data sets that reduces to linear complexity compared to quadratic complexity of dtw. The train data should be a numpy array of the form (M,N,D) where
- M - Number of data sequences.
- N - length of data sequences.
- D - Dimension of data sequences (number of features).
The algorithm was tested on a synthetic vehicle encounter dataset. Clustering vehicle encounter data into different kinds of encounters -
The dataset contained time sequences of 100 steps (duration of 10s) belonging to 3 different classes. Each class had 200 samples.
