Estimation of photometric redshifts is one of the essential tasks in cosmology, and in fact one of the precursor techniques for many analyses that require a measure of distance. These distances are given by the distance-redshift relation, and hence one needs very accurate measures of redshift to be confident in the inferred distances. Ideally high-resolution spectra would be obtained for every object which would allow for a precise measurement of the redshift. However, with current and future surveys such as the Dark Energy Survey (DES), Euclid, and the Large Synoptic Survey Telescope (LSST), hundreds of millions of galaxies will be observed, and even with new, large spectroscopic surveys such as the Dark Energy Spectroscopic Instrument (DESI), only a small fraction of the galaxies will have spectroscopy performed. Instead, photometric redshift estimates are required, particularly over large-scale datasets gathered from surveys. This project explores comparing a variety of techniques for the estimation of photometric redshift, including neural networks, random forests, linear regression and decision trees.
