Boundary-free Kernel-smoothed Goodness-of-fit Tests for Data on General Interval

TL;DR

This paper introduces boundary-free kernel-smoothed goodness-of-fit tests for data on general intervals using bijective transformations to address boundary issues in kernel-based methods, with simulation results demonstrating their effectiveness.

Contribution

It proposes a novel boundary-free approach for kernel-based goodness-of-fit tests on general intervals using bijective transformations.

Findings

Transformations effectively eliminate boundary problems.

Simulations show improved test performance.

Method outperforms naive kernel approaches.

Abstract

We propose kernel-type smoothed Kolmogorov-Smirnov and Cram\'{e}r-von Mises tests for data on general interval, using bijective transformations. Though not as severe as in the kernel density estimation, utilizing naive kernel method directly to those particular tests will result in boundary problem as well. This happens mostly because the value of the naive kernel distribution function estimator is still larger than $0$ (or less than $1$) when it is evaluated at the boundary points. This situation can increase the errors of the tests especially the second-type error. In this article, we use bijective transformations to eliminate the boundary problem. Some simulation results illustrating the estimator and the tests' performances will be presented in the last part of this article.