Kernel-based Optimally Weighted Conformal Time-Series Prediction

About

In this work, we present a novel conformal prediction method for time-series, which we call Kernel-based Optimally Weighted Conformal Prediction Intervals (KOWCPI). Specifically, KOWCPI adapts the classic Reweighted Nadaraya-Watson (RNW) estimator for quantile regression on dependent data and learns optimal data-adaptive weights. Theoretically, we tackle the challenge of establishing a conditional coverage guarantee for non-exchangeable data under strong mixing conditions on the non-conformity scores. We demonstrate the superior performance of KOWCPI on real and synthetic time-series data against state-of-the-art methods, where KOWCPI achieves narrower confidence intervals without losing coverage.

Jonghyeok Lee, Chen Xu, Yao Xie · 2024

Related benchmarks

Task	Dataset	Result
Time Series Forecasting Uncertainty Quantification	GIFT-Eval Bench22	nWink1.56	30
Time Series Forecasting Uncertainty Quantification	GIFT-Eval Bench10	nWink84	30
Sequential Conformal Prediction	Weather Rain	Win Rate64.44	24
Time Series Forecasting Uncertainty Quantification	GIFT-Eval Bench10_100k	nWink0.77	12
Time Series Forecasting Uncertainty Quantification	GIFT-Eval Bench10 full	nWink0.77	12
Sequential Conformal Prediction	META Stock	Win Rate61	8
Sequential Conformal Prediction	Solar Radiation	Win Rate2.07	8
Sequential Conformal Prediction	Wind Energy	Win Rate3.54	8
Sequential Conformal Prediction	Electricity	Win Rate2.44	8
Sequential Conformal Prediction	NVDA Stock	Win Rate0.72	8