Classification of Pulsar Candidates Using Classical Machine Learning Algorithms

Pulsars, rapidly spinning neutron stars with powerful radio emissions, are of significant interest in astrophysics. This study examines the effectiveness of various machine learning methods-Random Forest, XGBoost, Logistic Regression, Support Vector Machines (SVM), and Gradient Boosting-in identifying pulsars from a given dataset. We addressed the challenge of data imbalance, where pulsars were under-represented, resulting in an accuracy increase from 82.4% to 93.7% after correction. Additionally, we removed higher-order central moments from the data and found that this had no statistically significant impact on prediction accuracy. Our analysis of feature importance revealed that attributes from the Integrated Profile are more crucial for identifying pulsars than those from the DM-SNR curve.