Pulsar Star Detection: A Comparative Analysis of Classification Algorithms using SMOTE

Abstract

A Pulsar is a highly magnetized rotating compact star whose magnetic poles emit beams of radiation. The application of pulsar stars has a great application in the field of astronomical study. Applications like the existence of gravitational radiation can be indirectly confirmed from the observation of pulsars in a binary neutron star system. Therefore, the identification of pulsars is necessary for the study of gravitational waves and general relativity. Detection of pulsars in the universe can help research in the field of astrophysics. At present, there are millions of pulsar candidates present to be searched. Machine learning techniques can help detect pulsars from such a large number of candidates. The paper discusses nine common classification algorithms for the prediction of pulsar stars and then compares their performances using various classification metrics such as classification accuracy, precision and recall value, ROC score and f-score on both balanced and unbalanced data. SMOTE-technique is used to balance the data for better results. Among the nine algorithms, XGBoosting algorithm achieved the best results. The paper is concluded with prospects of machine learning for pulsar detection in the field of astronomy.