Machine Learning Classification to Identify Catastrophic Outlier Photometric Redshift Estimates
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.