Exact and Robust Conformal Inference Methods for Predictive Machine Learning With Dependent Data
This work addresses the challenge of reliable uncertainty quantification in predictive machine learning for time-dependent data, such as in finance or climate modeling, representing an incremental extension of conformal inference to non-exchangeable settings.
The authors tackled the problem of extending conformal inference to time series data with serial dependence, developing a method that retains exact validity under exchangeability and approximate validity under weak assumptions, achieving robust predictive intervals without requiring specific model assumptions.
We extend conformal inference to general settings that allow for time series data. Our proposal is developed as a randomization method and accounts for potential serial dependence by including block structures in the permutation scheme. As a result, the proposed method retains the exact, model-free validity when the data are i.i.d. or more generally exchangeable, similar to usual conformal inference methods. When exchangeability fails, as is the case for common time series data, the proposed approach is approximately valid under weak assumptions on the conformity score.