Comparison of machine learning algorithms and feature importance analysis for star classification

Abstract

This study presents a comprehensive investigation of star classification based on the Stellar Classification Dataset-SDSS17, employing machine learning algorithms including Random Forest, Gradient Boosting, and Support Vector Machine (SVM), along with Shapley Additive Explanations (SHAP) feature importance analysis. The research found that among the 17 features studied, redshift consistently emerged as the most significant feature. Additionally, the feature importance or SHAP value of Redshift obtained in various models is significantly higher than that of other features. The angles of Right Ascension (alpha) and Declination (delta), in contrast, showed the least importance. Models with higher accuracy tend to exhibit lower importance for Redshift. For the classifier result, Random Forest yielded the highest accuracy and SVM had the lowest accuracy. Most models perform best when classifying the "star" class and worst when classifying the "quasar" class. These findings provide valuable insights for automated star classification and underscore the critical role of redshift, thereby aligning with astronomical theories. Further research could include investigating more sophisticated models, like neural networks, and conducting a more profound analysis of feature interactions.