The effectiveness of different machine learning algorithms in classifying pulsar stars and the impact of data preparation

Abstract

With massive improvements in technology, astronomers have been uncovering impressive findings of the mysteries of the universe. At the same time, new technology is able to do more and more of the tasks usually given to people, especially when it comes to handling big pieces of data. As such, more jobs are being replaced by machines in the modern industrial age, which may very well include the jobs of astronomers. To investigate the potential of machine learning in astronomy and its effect on the livelihoods of astronomers, this paper aims to demonstrate the capabilities of machine learning algorithms in completing tasks usually done by astronomers, which in this case the identifying pulsar stars. Astronomers study these stars extensively to learn about extreme states of matter and explore different solar systems. This paper introduces various classification algorithms to perform the task of identifying pulsar stars from data of radio wave emissions in the universe. Two distinct methods of dealing with missing values were also used. Once the algorithms were applied to the dataset, the results were recorded and analyzed, mainly focusing on the accuracy of each model. From the results, all the models yielded accuracies above 90%. In addition, the models performed better on average with data preparation using the drop method compared to the average value method. These results provide extensive evidence that machine learning can perform certain tasks commonly performed by astronomers themselves. At the moment they cannot replace astronomers, but in the future, it has a possibility.