Conformal Statistical Inference: A Tutorial for Biostatistics and Machine Learning Applications

This tutorial introduces Conformal Statistical Inference — a distribution-free framework for constructing prediction sets with valid finite-sample coverage guarantees. It covers:

• Foundations: exchangeability, nonconformity scores, coverage probability guarantees
• Split Conformal Prediction (SCP): computationally efficient, data-splitting approach
• Conformal Quantile Regression (CQR): adaptive intervals handling heteroscedasticity
• Full Conformal Prediction: exhaustive leave-one-out calibration
• Jackknife+: conservative leave-one-out conformal with improved properties

Resources

• 🎓 Tutorial Slides (Spanish): Quarto RevealJS slides
• 📘 Comprehensive Guide (English): Interactive notebook
• 💻 Reproducible code available on GitHub