Dynamic Ensemble Selection with XGBoost for Enhanced Survival Analysis of Kidney Transplant Outcomes Using NMR Metabolomics

Akbari, Mortaza (2025) Dynamic Ensemble Selection with XGBoost for Enhanced Survival Analysis of Kidney Transplant Outcomes Using NMR Metabolomics. Master's Thesis / Essay, Artificial Intelligence.

Preview	Text mAI2025AkbariA.pdf Download (2MB) | Preview
	Text toestemming akbari.pdf Restricted to Registered users only Download (260kB)

Abstract

Survival analysis underpins time-to-event prediction in medicine, where accurate timing matters for decisions and outcomes. In kidney transplantation, anticipating graft failure can guide monitoring, immunosuppression, and re-listing. Classical tools such as the Cox proportional hazards model struggle when relationships are nonlinear and data are high-dimensional or censored. In this context, machine learning methods, used to extract predictive signal from complex measurements, offer a pragmatic way to interpret rich sources such as Nuclear Magnetic Resonance (NMR) metabolomics. This thesis evaluates dynamic ensemble selection (DES) for survival prediction, using XGBoost models ensembles to convert complex input data into clinically useful risk estimates. DES tailors, per patient, which models are combined - in contrast to static selection or uniform averaging - and is designed to exploit local structure in heterogeneous populations. We study two settings. First, a high-dimensional NMR cohort of kidney transplant recipients (n=249; 30 events) serves as the primary test bed. Second, as a cross-dataset check, we apply the same pipeline to NHANES with linked mortality follow-up (lower-dimensional, larger sample). The concordance index (C index) is the primary metric. In the NMR cohort, the DES strategy that prunes and then weights models (DES-WS) outperformed static selection and uniform averaging in the external test set (peak C-index ¡0.05), while single-model selection (DES-S

Item Type:	Thesis (Master's Thesis / Essay)
Supervisor name:	Schippers, M.B. and Guo, J.
Degree programme:	Artificial Intelligence
Thesis type:	Master's Thesis / Essay
Language:	English
Date Deposited:	04 Sep 2025 09:23
Last Modified:	04 Sep 2025 09:23
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/36972

Actions (login required)

View Item