scikit-survival

Survival analysis with censored time-to-event data in Python can be performed using the scikit-survival toolkit, which models censored records via Cox proportional hazards, ensemble methods, and survival SVMs.

Scikit-learn estimators can be used for survival analysis and competing risks through the scikit-survival library, which provides compatible wrappers for survival SVMs, RandomSurvivalForest, and cause-specific hazard modeling.

The best way to evaluate a Cox proportional hazards model in Python is using survival-specific metrics like the concordance index, time-dependent AUC, and Brier score, all integrated into the scikit-survival evaluation workflows.

Scikit-survival supports Random Survival Forests and penalized Cox models, offering CoxnetSurvivalAnalysis for elastic-net penalization alongside ensemble methods like ExtraSurvivalTrees and GradientBoostingSurvivalAnalysis.

Data preprocessing and loading for time-to-event modeling is handled through scikit-survival utilities, allowing you to load built-in datasets like GBSG2 and transform user-provided survival data into the required censored format.

Survival SVMs should be used instead of Cox models for time-to-event data when you need non-linear decision boundaries, utilizing FastSurvivalSVM or FastKernelSurvivalSVM rather than relying on proportional hazard assumptions.