Applying Regression Conformal Prediction with Nearest Neighbors to time series data

TL;DR

This paper explores adapting conformal prediction to time series data by using a nearest neighbors approach to construct reliable prediction intervals despite the data's dependence structure.

Contribution

It introduces a method combining conformal prediction with weighted nearest neighbors to handle the dependence in time series data, which challenges traditional conformal assumptions.

Findings

The proposed approach produces valid prediction intervals for time series.

The method demonstrates effectiveness on real datasets.

It addresses the dependency issue in conformal prediction for time series.

Abstract

In this paper, we apply conformal prediction to time series data. Conformal prediction isa method that produces predictive regions given a confidence level. The regions outputs arealways valid under the exchangeability assumption. However, this assumption does not holdfor the time series data because there is a link among past, current, and future observations.Consequently, the challenge of applying conformal predictors to the problem of time seriesdata lies in the fact that observations of a time series are dependent and therefore do notmeet the exchangeability assumption. This paper aims to present a way of constructingreliable prediction intervals by using conformal predictors in the context of time series. Weuse the nearest neighbors method based on the fast parameters tuning technique in theweighted nearest neighbors (FPTO-WNN) approach as the underlying algorithm. Dataanalysis demonstrates the effectiveness of the proposed approach.