Back to search

Athens University Economics and Business (AUEB)

Bayesian variable selection for survival analysis models

Abstract

dc:description

Selecting among different models has been gaining significant scientific concern over the last decades. This stems from the fact that due to the increased complexity of natural phenomena, the ability to understand and make sense of them is crucially important since this implies understanding the underlying mechanism that produces that phenomena and hence implies understanding and capturing the uncertainty behind them. Models that attempt to capture the uncertainty of natural phenomena can be drawn from a wide pool of distributions that can be both simple and complex depending on the problem under study. For a wide variety of problems, the Normal distribution has been used a default option due to its lack of complexity, the thorough understanding of its behavior, and the fact that it can be approximated under the Central Limit Theorem (CLT) provided a large enough sample. Unfortunately, such distributions, can be a useful modeling tool only for a small class of phenomena leading to the need of more complex models to explain real world issues or concerns. Among the real world issues is the modeling of time-to-event data. These recorded times can be of general interest, from modeling sport events (goal arrival times) to modeling medically related times (time until an active treatment is effective, or time until the death of a patient). These problems, by nature, entail complexity especially due to the wide variety of factors that can affect the outcome variable, the fact that the response (time-to-event) is strictly non negative and due to the existence of censoring. In this thesis, we focus on the class of Accelerated Failure Time (AFT) survival models as the means of the likelihood of these data, while providing default priors for objective Bayesian variable selection problems that to complete the Bayesian setup of modeling such data. More specifically, we focus on modeling data using the Weibull distribution while also starting the conversation regarding how this methodology can be extended by using more general distributions. The proposed prior has been applied in a wide variety of simulated data to fully understand its general and limiting behavior. Finally, the model selection consistency property has been demonstrated through simulated data.

Degree

thesis:*
Grantor dc:publisher
Athens University Economics and Business (AUEB)
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Leriou, Ilias
  • Λερίου, Ηλίας

Subjects

dc:subject × 8

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/61206

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Leriou, Ilias; Λερίου, Ηλίας. Bayesian variable selection for survival analysis models. Athens University Economics and Business (AUEB), 2025. http://hdl.handle.net/10442/hedi/61206