Machine Learning: Quality Control of HST Grism Spectra

The Pipeline for Hubble Legacy Archive Grism data (PHLAG) had been used to extract more than 70000 wavelength and flux calibrated ID spectra. They were obtained from 153 fields observed in G800L grism spectroscopy mode with the Advanced Camera for Surveys on the Hubble Space Telescope. This number of spectra is far too large to allow detailed visual inspection for quality control on reasonable time-scales. As a solution, we use machine learning techniques to classify spectra into "good" and "bad" based on a careful visual inspection of only about 3% of the full sample. A final visual skim through the set of "good" spectra was made to remove catastrophic failures. The remaining 47919 spectra form the largest set of slitless high-level spectroscopic data products publicly released to date.