A Comparative Study of Binary Classification Methods for Pulsar Detection

Pulsars are fast spinning neutron stars which on observing, emit pulsed appearance of radio waves and other electromagnetic radiation with very high pulse rate. The study of these dense neutron stars provides key insights on various physical occurrences like the plasma behavior in highly dense environments, behaviors of binary system consisting of a pulsar and a black hole and general relativity for the same. This requires a very elaborate dataset of pulsars and their statistical data for both repeatability and experimental accuracy. For this to be implemented, many large-scale pulsar surveys are conducted from time to time. During the process of the survey, manual classification of the data thus obtained, introduces bottleneck both in terms of labor needed and accuracy of classification. Hence statistical learning approaches can be used for the same for autonomous detection of pulsars. The raw dataset obtained for sampling is usually highly unbalanced and this study explores the comparison between the methods for diminishing the effects of unbalanced training datasets on different supervised classifiers to increase the accuracy of classification.