CLASSIFICATION OF MULTIVARIATE TIME SERIES USING BAYESIAN DYNAMIC MODELS

Consider a finite set of classes of discrete-time stochastic processes, each of them with unknown specifications and parameters. Given an observed time series, whose class is unknown, the time series classification problem deals with identifying which class the observed series belongs to. In this article, the online classification of multivariate series is addressed, in which the observed series is classified as its observations are obtained. A model composed of an ensemble of Bayesian dynamic linear models is proposed, in which each model is adjusted to one of the classes based on samples. Given a series whose class is unknown, the probability of each class is computed using Bayes' theorem. The model is applied to the problem of recognizing human activity using inertial sensors, obtaining 96% accuracy in classifying activities.