Machine Learning Classification to Identify Catastrophic Outlier Photometric Redshift Estimates

TL;DR

This paper demonstrates that a simple neural network classifier can effectively identify galaxies with catastrophic outlier photometric redshift estimates using only their photometric magnitudes, reducing errors in scientific analyses.

Contribution

The study introduces a basic neural network approach for outlier detection in photometric redshifts, highlighting its effectiveness with minimal complexity.

Findings

Significant fraction of outliers identified

Low false positive rate for non-outliers

Potential to improve redshift-based analyses

Abstract

We present results of using a basic binary classification neural network model to identify likely catastrophic outlier photometric redshift estimates of individual galaxies, based only on the galaxies' measured photometric band magnitude values. We find that a simple implementation of this classification can identify a significant fraction of galaxies with catastrophic outlier photometric redshift estimates while falsely categorizing only a much smaller fraction of non-outliers. These methods have the potential to reduce the errors introduced into science analyses by catastrophic outlier photometric redshift estimates.