Title: A new classification method for multivariate time series data
Authors: Soudeep Deb - Indian Institue of Management Bangalore (India) [presenting]
Shubhajit Sen - North Carolina State University (United States)
Abstract: Classification of multivariate time series (MTS) data has applications in various domains, for example, medical sciences, finance, sports analytics, etc. Though the classification of univariate time series (UTS) is a well-explored area, unfortunately, the same cannot be said for MTS classification. We propose a new technique that uses the advantages of dimension reduction through the t-distributed stochastic neighbor embedding (t-SNE) method, coupled with the attractive properties of the spectral density estimates of a time series, and k-nearest neighbor (k-NN) algorithm. We transform each MTS to a lower-dimensional time series, making it useful for visualizing and retaining the temporal patterns, and subsequently use that in classification. Then, we extend the standard univariate spectral density-based classification in the multivariate setting and prove its theoretical consistency. For real-life data analysis, we have chosen two health-related datasets. Empirically, at first, we establish that the pair-wise structure of the multivariate spectral density-based distance matrix is retained in the t-SNE transformed spectral density-based method. Then, for both cases, the proposed algorithm is implemented and we find that it achieves much better classification accuracy than the other widely used methods.