Computation of the Z parameter of galaxies from astronomical data bases using deep learning techniques

Abstract

This dissertation is a continuation of Apostolos Kiraleos’ work on predicting galaxy redshift. We will use a dataset provided by the Eu- ropean Space Agency’s ”Gaia Mission” and we will try to implement a neural network that predicts a galaxy’s redshift using its spectral data. Most importantly, we will focus on finding ways to predict a galaxy’s redshift on extreme values. More specifically, we will delve deeply into the problem of imbalanced regression using different tactics and other network architecture designs. Finally, we will evaluate the final model by measuring its performance mainly on extreme values and outliers, then we will compare the results with Kiraleos’ model. The solutions provided are important not only in the field of astronomy but also in every domain of science that handles imbalanced datasets.