Implementation of machine learning algorithms in the Sloan Digital Sky Survey DR14 analysis

Abstract

Abstract The fourth edition of the Sloan Digital Sky Survey has been investigated in the paper. There are a few telescopes analyzing sky at different frequencies. They generate a lot of statistical data combined into datasets. One of them is explored in the paper. The handled dataset contains information about three types of objects: stars, quasars and galaxies. Efforts of physicists aren’t enough to investigate vast amount of data. The goal of machine learning implemented in this area is to solve the most tasks of classification in automatical way. Attention should be paid only to some complicated cases. Information in such datasets is already marked up in order to apply classification algorithms and models. Review of literature has shown that neural networks are often used to investigate such datasets that could be handled with simple models. In this research some simple classification models are implemented, as well there are results of ensemble algorithms implementation. Advantages and disadvantages of their implementations are described, physical explanation of classifiers’ structure is presented when it’s possible. Results and conclusions could be used in processing of other astronomical datasets.