Robust Chauvenet Rejection: Powerful, but Easy to Use Outlier Detection for Heavily Contaminated Data Sets

TL;DR

This paper introduces a Python implementation of Robust Chauvenet Rejection (RCR), an effective outlier detection method for heavily contaminated data sets, demonstrating its accuracy, speed, and versatility in one- and multi-dimensional contexts.

Contribution

The paper presents a new Python package for RCR, enhancing its accessibility and usability while maintaining its high performance for outlier rejection in contaminated data.

Findings

RCR effectively cleans heavily contaminated data sets.

The Python implementation maintains the speed of the original C++ version.

RCR performs well in both one-dimensional and multi-dimensional data analysis.

Abstract

In Maples et al. (2018) we introduced Robust Chauvenet Outlier Rejection, or RCR, a novel outlier rejection technique that evolves Chauvenet's Criterion by sequentially applying different measures of central tendency and empirically determining the rejective sigma value. RCR is especially powerful for cleaning heavily-contaminated samples, and unlike other methods such as sigma clipping, it manages to be both accurate and precise when characterizing the underlying uncontaminated distributions of data sets, by using decreasingly robust but increasingly precise statistics in sequence. For this work, we present RCR from a software standpoint, newly implemented as a Python package while maintaining the speed of the C++ original. RCR has been well-tested, calibrated and simulated, and it can be used for both one-dimensional outlier rejection and $n$-dimensional model-fitting, with or without weighted data. RCR is free to use for academic and non-commercial purposes, and the code, documentation and accompanying web calculator can be found and easily used online at https://github.com/nickk124/RCR