scikit-survival

You can train a survival model on censored time-to-event data in Python by creating Surv outcomes and fitting an estimator like Cox proportional hazards, Random Survival Forest, or survival SVMs using scikit-survival to handle censored observations directly.

Survival analysis models with censored data are evaluated using censoring-aware metrics such as Harrell's or Uno's concordance index, time-dependent AUC, and Brier or Integrated Brier scores to assess discrimination and calibration.

Yes, you can estimate cumulative incidence functions for competing risks analysis by using dedicated non-parametric utilities that handle competing event types within the scikit-survival framework.

To compare Cox proportional hazards and Random Survival Forests, fit both estimators on your censored dataset and evaluate their discrimination using Uno's C-index to determine which model predicts event risk more accurately.

Yes, scikit-survival supports Gradient Boosting survival models and survival SVMs, allowing you to fit ensemble and margin-based estimators on censored time-to-event datasets for predicting event risk.