A segment-wise extraction of multivariate time-series features for Grassmann clustering

In this paper, a novel approach of extracting features from multivariate time-series (MTS) with different time lengths, is proposed to enhance the clustering accuracy. Particularly, the feature extraction is conducted on time-sample segments of MTS, in which several segments are defined without overlapping. As for feature extractor, the conventional two-dimensional principal component analysis (2DPCA) is deployed due to its proven effectiveness in feature representation. Our experimental results show that the proposed segment-wise extraction of 2DPCA features is helpful in enhancing the clustering accuracy.