Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

Author ORCID Identifier



Open Access Dissertation

Document Type


Degree Name

Doctor of Philosophy (PhD)

Degree Program


Year Degree Awarded


Month Degree Awarded


First Advisor

John Staudenmayer

Second Advisor

Krista Gile

Third Advisor

Michael Lavine

Fourth Advisor

Patty Freedson

Subject Categories



In this thesis we develop methods for classifying physical activity using accelerometer recordings. We cast this as a problem of classification in time series with moderate to high dimensional observations at each time point. Specifically, we observe a vector of summary statistics of the accelerometer signal at each point in time, and we wish to use these observations to estimate the type and intensity of physical activity the individual engaged in as it changes over time.

Our methods are based on Conditional Random Fields, which allow us to capture temporal dependence in an individual’s physical activity type without requiring us to model the distribution of the observed features at each point in time. We develop three novel estimation strategies for Conditional Random Fields, evaluate their performance on classification tasks through simulation studies and demonstrate their use in applications with real physical activity data sets.


Included in

Biostatistics Commons