Prediction of the Stellar Class of a Star Based on Its Characteristics Using Machine Learning

We present techniques for a machine learning approach to predict stellar classes of stars using their physical and observable characteristics. Stellar classification is the categorization of stars into spectral types and subclasses based on temperature, color, and various other properties, and it is a fundamental aspect of astronomy, providing vital insights into stellar properties and evolutionary stages. By taking the general class of a star (e.g., G2 being the Sun), our model leverages a diverse range of input features, such as luminosity, surface temperature, and color indices, to predict a star’s spectral class with viable accuracy. Among several implemented algorithms, Random Forest Classifier achieved an accuracy of 76% (log loss 0.69), outperforming other methods such as XGBoost (71%), K-Nearest-Neighbors (50%), and Logistic Regression (23%). We attribute the lower performance of XGBoost to overlapping threshold features and K-Nearest-Neighbors to the low linear correlations of the data. The results demonstrate the high potential of machine learning to automate feature classification within astronomy, such as spectral classification efficiently and with high accuracy, significantly enhancing our capacity to analyze large data sets in modern astronomy.