Conformal Prediction Methods in Uncertainty Quantification

Conformal prediction provides a nonparametric framework for generating statistically valid prediction intervals or sets under minimal assumptions. At its core is the calibration of model outputs through a nonconformity measure, which ranks the “strangeness” of new observations relative to a reference set. Under the assumption of exchangeability, conformal methods guarantee that their predictive regions contain the true outcome with a user-specified frequency in finite samples. Multiple variants have been developed: transductive conformal prediction recalibrates on each test point, inductive conformal prediction improves efficiency through a separate calibration set, and conformalized quantile regression adapts to heteroscedastic noise by combining quantile estimates with conformal adjustments. Extensions encompass conformal predictive distributions, which yield full probability forecasts, and conformal e-prediction, which operates via e-values rather than p-values. Recent methodological advances include cluster-based adaptation to local variability, jackknife and cross-conformal techniques for enhanced coverage, and homotopy methods to enable full conformal inference in high-dimensional settings. Applications span forest attribute mapping, survival analysis, building energy forecasting and complex interaction models, emphasising the global significance of robust, distribution-free uncertainty quantification in science and engineering.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Innovations in forest attribute mapping have demonstrated the use of conformal k-nearest neighbours procedures to produce valid prediction intervals for metrics such as tree volume and height across diverse forest strata. These approaches, based on quantiles or standard deviations of nearest neighbours, achieve theoretical coverage guarantees and adapt to spatial heterogeneity while remaining competitive with quantile regression methods.

In the context of energy systems, conformal prediction has been applied to building performance simulation to deliver model-agnostic and assumption-free uncertainty estimates for heating and cooling load forecasting. By wrapping any pre-trained predictor in a conformal calibration step, researchers have produced reliable prediction intervals that support informed decision-making in energy design and operation, irrespective of data distribution assumptions.

Conformal techniques have also advanced survival analysis by generating lower predictive bounds on time-to-event outcomes under right-censoring. This covariate-dependent method provides finite-sample validity without strong parametric assumptions and exhibits a doubly robust property under conditional censoring. Empirical studies on synthetic and clinical data confirm its ability to yield informative and calibrated predictions in high-stakes medical settings.

Conformal Prediction Methods in Uncertainty Quantification publication trend

The graph below shows the total number of articles in conformal prediction methods in uncertainty quantification across all publications each year (not limited to Nature Index journals).

Technical terms

Exchangeability: A property of a data sequence indicating that the joint probability distribution is invariant under permutation, underpinning the validity of conformal methods.

Nonconformity measure: A function that quantifies how atypical a new observation is relative to a reference set, guiding the construction of prediction regions.

Inductive conformal prediction: A scalable variant that splits data into training and calibration subsets, computing prediction intervals without retraining on each test point.

Prediction interval: A set of values within which a future observation is expected to lie with a specified probability, guaranteed by conformal methods.

References

1. Uncertainty quantification for forest attribute maps with conformal prediction and k -nearest neighbor method. Remote Sensing of Environment (2025).
2. Quantifying Uncertainty with Conformal Prediction for Heating and Cooling Load Forecasting in Building Performance Simulation. Energies (2024).
3. Conformalized survival analysis. Journal of the Royal Statistical Society Series B Statistical Methodology (2023).
4. Nonparametric predictive distributions based on conformal prediction. Machine Learning (2018).
5. Conformal e-prediction. Pattern Recognition (2025).
6. Improving conformalized quantile regression through cluster-based feature relevance. Expert Systems with Applications (2024).
7. Probabilistic prediction with locally weighted jackknife predictive system. Complex & Intelligent Systems (2023).
8. A confidence machine for sparse high‐order interaction model. Stat (2024).