Detection of Pulsars Using an Artificial Neural Network

The research paper demonstrates how to devise an optimal machine learning classifier for the detection of pulsars and then analyzes the performance of various classification models. Pulsars are classified as zero (not a pulsar) or one (pulsar) using logistic regression, decision tree, random forests, KNN classifier, and an artificial neural network. The performance has been analyzed based on five parameters: accuracy, recall, precision, specificity, and prevalence. For this, emission data from the HTRU2 dataset which was collected from the High Time Resolution Universe Survey is being used. Before using any classification algorithms, the patterns in the data were analyzed to understand the correlation between the different characteristics. The classifier proposed in this paper was found to give 98.01% accuracy.