Please use this identifier to cite or link to this item:
http://hdl.handle.net/10609/146174
Title: | Regresión logística multitarea para análisis de supervivencia: de los modelos tradicionales a los de aprendizaje automático |
Author: | Vallarino Navarro, Diego |
Tutor: | Perez-Alvarez, Nuria |
Others: | Ventura Arroyo, Carles |
Abstract: | In the present work we have used a real database, from the survival package, to be able to test if there was an improvement in performance in the use of different survival models. After making a conceptual discussion about four models, a parametric model, a semiparametric model, a non-parametric model, and another within the category of machine learning, we have shown that the models have different performances. Possibly the answer to this improvement in performance lies in the use of censored data differently within the development of each model, as evidenced in the theory analysed in this paper. We base the previous hypothesis on the fact that the model that has the best performance, measured by the C-index, is the multitask logistic regression (MTLR) model, which is essentially a collection of logistic regression models built at different time intervals. To determine the probability that the event of interest would occur during each interval. The results provided by the MTLR are similar to the CoxPH model without relying on the CoxPH assumption that the hazard function for the two subjects is constant over time. The performance improvement of the MTLR over the Cox model, the closest in performance, was approximately 6%. |
Keywords: | survival analysis machine learning database |
Document type: | info:eu-repo/semantics/masterThesis |
Issue Date: | 1-Jun-2022 |
Publication license: | http://creativecommons.org/licenses/by-nc-nd/3.0/es/ |
Appears in Collections: | Trabajos finales de carrera, trabajos de investigación, etc. |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
dvallarinoTFM0622memoria.pdf | Memoria del TFM | 933,55 kB | Adobe PDF | View/Open |
Share:
This item is licensed under a Creative Commons License