Reproduce “Optimal Survival Trees” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dimitris Bertsimas, Jack Dunn, Emma Gibson, Agni Orfanoudaki
Optimal Survival Trees (OST), developed at MIT's Operations Research Center, applies mixed-integer optimization to build globally or near-globally optimal decision trees for survival analysis with censored data. This method consistently achieves superior accuracy and stability over traditional greedy survival tree algorithms across simulated and real-world datasets, offering interpretable models for predicting event times.
Key insights
- Leveraging global mixed-integer optimization allows for the construction of single survival trees that are both highly accurate and fully interpretable, addressing the traditional trade-off between these objectives.
- A carefully designed survival-specific loss function, satisfying a 'node purity' criterion, is critical for integrating censored data into a computationally efficient global optimization framework.
- Novel simulation-specific metrics (Node Homogeneity, Class Recovery, Kaplan-Meier Area Ratio) provide a more comprehensive evaluation of structural and predictive accuracy for survival trees than standard survival metrics alone.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.