Reliable Prediction Intervals for Local Linear Regression

TL;DR

This paper presents BOPI, a new method for reliable prediction intervals in local linear regression, validated through experiments and simulations showing improved coverage and size compared to existing methods.

Contribution

Introduces BOPI for prediction intervals and a new comparison measure EGSD, with extensive validation demonstrating its effectiveness.

Findings

BOPI achieves better coverage probability.

BOPI produces smaller mean interval sizes.

EGSD effectively compares interval prediction models.

Abstract

This paper introduces two methods for estimating reliable prediction intervals for local linear least-squares regressions, named Bounded Oscillation Prediction Intervals (BOPI). It also proposes a new measure for comparing interval prediction models named Equivalent Gaussian Standard Deviation (EGSD). The experimental results compare BOPI to other methods using coverage probability, Mean Interval Size and the introduced EGSD measure. The results were generally in favor of the BOPI on considered benchmark regression datasets. It also, reports simulation studies validating the BOPI method's reliability.