Multivariate Methods for Classifying Physiological Data

Abstract

In this paper we examine two novel multivariate time series representations to classify physiological data of different lengths, Multivariate Bag-of-Patterns and Stacked Bagsof-Patterns. We also borrow techniques from the natural language processing and text mining (e.g., term frequency and inverse document frequency) to improve classification accuracy. We compare how these multivariate representations classify the data compared to extensions of two univariate representations, known as Piecewise Dynamic Time Warping and Bag-of-Patterns, into the multivariate domain. We present experimental results on classifying adult patients who have experienced acute episodes of hypotension (AHE) and neonatal patients who have experienced a patent ductus arteriosus (PDA). We also evaluated how these methods fared in classifying robotic sensor data to determine location and direction of the robot and motion capture data to differentiate types of motions to determine whether the methods are generalizable to other domains.