Duwinanto, Muhammad Rizki (2025) A Custom Distribution Fitted Random Forest for Accelerated Failure Time Modelling. Master's Thesis / Essay, Artificial Intelligence.
|
Text
mAI2025DuwinantoMR.pdf Download (1MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
Toestemming.pdf Restricted to Registered users only Download (277kB) |
Abstract
Survival Analysis is crucial in healthcare for predicting the occurrence of a specific event in a population, such as a patient’s death or a particular outcome. Most survival analysis methods are currently focused on predicting the survival risk of an event within a certain time, using Random Survival Forest and Neural Networks. To predict the occurrence of time-to-event directly, the Accelerated Failure Time model can be used. Existing methods, such as XGBoost AFT, have already been explored but lack a more data-driven approach. In this research, a random forest is implemented using the Accelerated Failure Time (AFT) loss function to explore the custom-distribution fitting. This research aims to improve the accelerated failure loss function using parametric fitting for three distributions in logarithmic time space of Logistic, Extreme and Normal and non-parametric fitting using Gaussian Mixture Models before the training. Finally, the bootstrapping method is used to avoid overfitting during the distribution fitting process. Custom distribution fitting dramatically enhanced the C-index and metrics, but bootstrapping yielded no better results. The usage of Gaussian mixture models also creates worse training and testing performances. In conclusion, Accelerated Failure Random Forest represents a promising step toward more effective direct time-to-event prediction models.
| 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: | 12 Sep 2025 08:53 |
| Last Modified: | 12 Sep 2025 08:53 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/37028 |
Actions (login required)
 |
View Item |