Leveraging Redshift Probabilities and Machine Learning Classification to Identify Catastrophic Outlier Photometric Redshift Estimates

Abstract We present results of using a basic binary classification neural network model to identify likely catastrophic outlier (CO) photometric redshift estimates of individual galaxies. The classification is based on individual galaxy redshift probability distributions generated by a multiclass classifier neural network and a support vector machine. Importantly, the redshift probability distributions used as input to the binary classifier can be used for photometric redshift estimates generated by a different method. We find that a simple implementation of this classification can identify a very large fraction of galaxies with CO photometric redshift estimates while erroneously categorizing only a much smaller, in some cases negligible, fraction of nonoutliers. These methods can reduce the errors introduced into science analyses by CO photometric redshift estimates.