What problem does it solve?
scikit-survival helps you perform time-to-event (survival) analysis when outcomes are censored, so you can model how features affect event risk without discarding incomplete observations.
Core Features & Use Cases
- Fitting survival models for censored data: Build Cox proportional hazards models (standard and penalized), accelerated failure time variants, and survival SVMs.
- High-performance non-linear modeling: Use ensemble methods such as Random Survival Forests and Gradient Boosting for complex feature–time relationships.
- Evaluating survival predictions correctly: Measure discrimination and calibration using Uno/Harrell C-index variants, time-dependent AUC, and Brier/Integrated Brier scores.
- Handling competing risks: Estimate cumulative incidence functions for mutually exclusive event types and avoid incorrect Kaplan-Meier interpretations.
- Typical use case: You have a clinical dataset with right-censoring and covariates (e.g., age, stage, biomarkers) and you need to compare multiple survival models and select the best-performing approach using IPCW C-index and Integrated Brier Score.
Quick Start
Use the scikit-survival skill to fit a Cox model on your structured survival dataset created with sksurv.util.Surv and then evaluate it using Uno’s concordance index.
