Back to search

Cornell University

Methods for Complex Time Series Data: Holistic Bayesian Changepoint Analysis and Fair Generative Modeling

Abstract

dc:description.abstract

Changepoint detection and time series analysis are vital for applications across finance, biomedicine, and sociology. Existing methods often assume uniform sampling or simplified data-generating mechanisms, and typically return a single set of changepoints, offering limited insight into uncertainty. This dissertation introduces flexible Bayesian and generative frameworks to overcome these limitations. In Section 1, we propose a Bayesian approach that treats partitions as random objects, enabling principled characterization of both changepoint presence and location uncertainty. Using a depth-based geometry for posterior partition distributions, we show how this resolves issues with thresholding MCMC draws, which fail to capture partition ordering or variability in changepoint locations. Simulations demonstrate improved detection of subtle changes, clearer separation of presence versus location uncertainty, and more stable inference compared to classical methods. Section 2 extends adaptive Bayesian trend filtering to irregularly sampled time series, accommodating heterogeneous temporal resolution and enabling realistic modeling of complex signals such as heart rate dynamics. Finally, in Section 3, we develop a generative-adversarial approach for fair classification that balances predictive performance with mitigation of group-level biases through the synthesis of realistic, demographically balanced training data.

Degree

thesis:*
Name thesis:degree_name
Ph. D., Statistics
Level thesis:degree_level
Doctor of Philosophy
Discipline thesis:degree_discipline
Statistics
Grantor
Cornell University
Year dc:date.issued
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Smits, Georgia
Chair dc:contributor.chair
  • Matteson, David
Committee members dc:contributor.committeemember
  • Basu, Sumanta
  • Wells, Martin

Subjects

dc:subject × 4

Rights

Language dc:language.iso
en

Identifiers

dc:identifier.*
Dc Identifier Bibid
17423428
Dc Identifier Other
ProQuest Submission ID: 15374
ProQuest Publication ID: 32395155

Chain of custody

source
Harvested from
Cornell University
Base URL
ecommons.cornell.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Smits, Georgia. Methods for Complex Time Series Data: Holistic Bayesian Changepoint Analysis and Fair Generative Modeling. Doctor of Philosophy thesis, Cornell University, 2025. https://hdl.handle.net/1813/121026