Photometric redshifts estimation for galaxies by using FOABP-RF

Abstract

ABSTRACT This paper proposes a new combinatorial algorithm (FOABP-RF)-using Fruit Fly Optimization Algorithm to enhance Back Propagation Neural Network (FOABP) and random forest (RF) to estimate photometric redshifts of galaxies. This method can improve the estimation accuracy and effectively overcome the shortcomings of artificial neural network which often falls into the local optimal point. And it is suitable for different types of galaxies. First, self-organizing feature mapping (SOM) is used to cluster samples into early-type and late-type galaxies. Then the Back Propagation neural network (BP), genetic algorithm and back propagation (GABP) neural network, particle swarm optimization algorithm combined with BP neural network (PSOBP), FOABP-RF and other latest algorithms are used to estimate the redshifts of the two types of galaxies from one to another. Finally, in the experiment, 80218 galaxies with the redshift Z < 0.8 from the Sloan Digital Sky Survey Data Release 13 (SDSS DR13) are used as the data set. The root mean squared error (RMSE) of early-type galaxies by FOABP-RF is 6.03, 2.41, and 1.94 per cent lower than BP, GABP, and PSOBP, respectively. And the RMSE of late-type galaxies by FOABP-RF is 6.09, 4.09, 73.37 per cent lower than BP, GABP, and PSOBP, respectively. This proves FOABP-RF is very suitable for estimating photometric redshifts.