Photometric redshifts for the S-PLUS Survey: Is machine learning up to the task?

The Southern Photometric Local Universe Survey (S-PLUS) is a novel project that aims to map the Hemisphere using twelve filter system, comprising five broad-band SDSS-like filters and seven narrow-band optimized for important stellar features in local universe. In this paper we use photometry morphological information from first S-PLUS data release (S-PLUS DR1) cross-matched unWISE spectroscopic redshifts Sloan Digital Sky DR15. We explore three different machine learning methods (Gaussian Processes with GPz two Deep Learning models made TensorFlow) compare them currently used template-fitting method DR1 address whether can take advantage of system photometric redshift prediction. Using tests accuracy both single-point estimates such as calculation scatter, bias, outlier fraction, probability distribution functions (PDFs) Probability Integral Transform (PIT), Continuous Ranked Score (CRPS) Odds distribution, conclude deep-learning combination Bayesian Neural Network Mixture Density offers most accurate current test sample. It achieves scatter ($\sigma_\text{NMAD}$) 0.023, normalized bias -0.001, fraction 0.64% galaxies r-auto magnitudes between 16 21.