Vector-valued (or multivariate) time series data commonly occur in various sciences. While modeling univariate time series is well-studied, modeling of multivariate time series, especially finite-valued or categorical, has been relatively unexplored. In this dissertation, we employ hidden Markov models (HMMs) to capture temporal and multivariate dependencies in the multivariate time series data. We modularize the process of building such models by separating the modeling of temporal dependence, multivariate dependence, and non-stationary behavior. We also propose new methods of modeling multivariate dependence for categorical and real-valued data while drawing parallels between these two seemingly different types of data. Since this work is in part motivated by the problem of prediction precipitation over geographic regions from the multiple weather stations, we present in detail models pertinent to this hydrological application and perform a thorough analysis of the models on data collected from a number of different geographic regions.
Cited By
- Fridman L, Reimer B, Mehler B and Freeman W Cognitive Load Estimation in the Wild Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems, (1-9)
- Baba A Removing Object Bouncing from Indoor Tracking Data Proceedings of the 2017 ACM Southeast Conference, (134-141)
- Baba A, Jaeger M, Lu H, Pedersen T, Ku W and Xie X Learning-Based Cleansing for Indoor RFID Data Proceedings of the 2016 International Conference on Management of Data, (925-936)
- Chounta I and Avouris N Time series analysis of collaborative activities Proceedings of the 18th international conference on Collaboration and Technology, (145-152)
- Taylor G and Hinton G Products of Hidden Markov Models Proceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence, (522-529)
Recommendations
Semiparametric hidden Markov models: identifiability and estimation
We review the theory on semiparametric hidden Markov models HMMs, that is, HMMs for which the state-dependent distributions are not fully parametrically, but rather semi- or nonparametrically specified. We start by reviewing identifiability in such ...
A copula approach for dependence modeling in multivariate nonparametric time series
AbstractThis paper is concerned with modeling the dependence structure of two (or more) time series in the presence of a (possibly multivariate) covariate which may include past values of the time series. We assume that the covariate ...
COPAR-multivariate time series modeling using the copula autoregressive model
The analysis of multivariate time series is a common problem in areas like finance and economics. The classical tools for this purpose are vector autoregressive models. These however are limited to the modeling of linear and symmetric dependence. We ...