Scikit Survival

소개

Provides a comprehensive toolkit for analyzing censored data and modeling the time until specific events occur, built directly on the familiar scikit-learn framework. It enables the implementation of diverse survival models—from traditional Cox proportional hazards to advanced machine learning ensembles like Random Survival Forests and Gradient Boosting—while offering specialized evaluation metrics such as Uno's C-index and Brier scores to ensure predictive accuracy. This skill is essential for researchers and data scientists working with clinical trials, customer churn analysis, or industrial reliability engineering where observations are often incomplete.

주요 기능

• Implementation of diverse survival models including Cox proportional hazards, Random Survival Forests, and Survival SVMs.
• 2,066 GitHub stars
• Comprehensive model evaluation using concordance indices, Brier scores, and time-dependent AUC.
• Support for high-dimensional feature selection and competing risks analysis in complex datasets.
• Estimation of non-parametric survival functions through Kaplan-Meier and Nelson-Aalen methods.
• Specialized handling of censored data using dedicated utilities for preparing time-to-event outcomes.

사용 사례

• Estimating the remaining useful life (RUL) and failure probabilities for industrial equipment.
• Predicting patient survival outcomes or disease recurrence in medical and clinical research.
• Analyzing customer retention and time-to-churn for subscription-based business models.