Predictive survival analysis with scikit-learn, scikit-survival and lifelines
summary
Predictive survival methods get specific: code runs in a JupyterLite notebook, the demo uses scikit-learn preprocessing, the lifelines Kaplan‑Meier estimator to get unbiased survival curves on 10,000 simulated trucks, and Cox models (with spline + kernel features) that raise C‑index above 0.5. scikit‑survival and survival forests fail to run in JupyterLite due to compiled deps, so an experimental hazardous gradient‑boosting incidence model is shown and outperforms Cox on IBS and ranking. Notebooks and data live on the linked GitHub.
AI-written summary · about
chapters (11)
- 0:00Survival analysis tools
- 3:38Audience setup poll
- 5:48Right-censored time-to-event data
- 17:39Naive regression fails with censoring
- 21:40Kaplan-Meier modeling
- 25:14Survival function fundamentals
- 37:46Model evaluation metrics
- 44:39Cox Proportional Hazards
- 1:08:54Feature engineering
- 1:17:44Gradient boosting for survival
- 1:22:00Summary and Q&A
related talks
01h 24mBayesianism and Survival Analysis
05mA crash-update to lifelines
35mUsing Survival Analysis to understand customer retention
27mUsing Machine Learning to solve a classification problem with scikit-learn
39mPredicting Titanic Survivors with Machine Learning
19mtidymodels for time-to-event data
01h 26mNew Trends in Estimation and Inference
54m"Would YOU Survive the Titanic?", with ML and .NET