Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ChauBoxplot and AdaptiveBoxplot: two R packages for boxplot-based outlier detection

Tiejun Tong, Hongmei Lin, Bowen Gang, Riquan Zhang

TL;DR

The paper tackles the problem that Tukey's boxplot with a fixed fence can overflag outliers as the sample size grows. It introduces two R packages, ChauBoxplot and AdaptiveBoxplot, implementing a Chauvenet-type fence with $k_n^{\rm Chau}$ and a data-adaptive, multiple-testing‑based pipeline (Holm for FWER and BH for FDR), respectively. Through simulations across diverse $n$ and a contaminated normal model, the authors show that Chauvenet-type and FDR-based boxplots maintain better control of false positives while retaining sensitivity to true outliers, whereas the fixed Tukey rule performs poorly at scale and FWER is overly conservative. The work provides practical guidance for selecting boxplot-based outlier methods across small to large datasets and offers CRAN-hosted, well-documented tools for reproducible, interpretable outlier detection.

Abstract

Tukey's boxplot is widely used for outlier detection; however, its classic fixed-fence rule tends to flag an excessive number of outliers as the sample size grows. To address this limitation, we introduce two new R packages, ChauBoxplot and AdaptiveBoxplot, which implement more robust methods for outlier detection. We also provide practical guidance, drawn from simulation results, to help practitioners choose suitable boxplot methods and balance interpretability with statistical reliability.

ChauBoxplot and AdaptiveBoxplot: two R packages for boxplot-based outlier detection

TL;DR

The paper tackles the problem that Tukey's boxplot with a fixed fence can overflag outliers as the sample size grows. It introduces two R packages, ChauBoxplot and AdaptiveBoxplot, implementing a Chauvenet-type fence with and a data-adaptive, multiple-testing‑based pipeline (Holm for FWER and BH for FDR), respectively. Through simulations across diverse and a contaminated normal model, the authors show that Chauvenet-type and FDR-based boxplots maintain better control of false positives while retaining sensitivity to true outliers, whereas the fixed Tukey rule performs poorly at scale and FWER is overly conservative. The work provides practical guidance for selecting boxplot-based outlier methods across small to large datasets and offers CRAN-hosted, well-documented tools for reproducible, interpretable outlier detection.

Abstract

Tukey's boxplot is widely used for outlier detection; however, its classic fixed-fence rule tends to flag an excessive number of outliers as the sample size grows. To address this limitation, we introduce two new R packages, ChauBoxplot and AdaptiveBoxplot, which implement more robust methods for outlier detection. We also provide practical guidance, drawn from simulation results, to help practitioners choose suitable boxplot methods and balance interpretability with statistical reliability.
Paper Structure (5 sections, 2 equations, 1 figure)

This paper contains 5 sections, 2 equations, 1 figure.

Table of Contents

  1. Introduction
  2. The ChauBoxplot Package
  3. The AdaptiveBoxplot Package
  4. Illustrating Example
  5. Conclusion and Practical Guidance

Figures (1)

  • Figure 1: Comparison of boxplot-based outlier detection methods for normal data with three contaminated outliers. Each panel displays the results for one method (Tukey, Chauvenet-type, FWER [Holm] with rate 0.05, or FDR [BH] with rate 0.05) at $n = 50$, $500$, $5000$ and $50\,000$, respectively.