Leveraging Machine Learning for Accurate Star Formation Rate Predictions with MAGPHYS

Abstract

Star formation rates (SFRs) are pivotal for understanding the growth of stars, galaxies, and the universe. Understanding SFR is essential for insights into galaxy evolution, stellar populations, cosmology, and interstellar dynamics. SFR analysis is well-suited to machine learning due to its complexity and volume of data. In our study, we utilized machine learning models on a dataset containing various factors such as gas luminosity, star formation timescale, and metallicity to predict SFRs. Our models included Linear Regression, Lasso Regression, and a neural network. Both Linear Regression and Lasso Regression yielded low mean squared error values, with the neural network achieving even lower values, demonstrating the superior performance of deep learning in determining SFRs. Additionally, we assessed feature importance for the Linear and Lasso Regression models, identifying which factors most significantly influence SFR predictions. From our analysis, we concluded that the aforementioned factors are crucial for accurately identifying SFRs in a galaxy, as our results showed that machine learning can predict SFRs with a mean squared error of 0.000939 and R-squared of 0.4808 based on galactic properties. Furthermore, we used graphs to illustrate the relationships between SFRs and different galactic properties, providing visual evidence of these connections. Our findings underscore the potential of machine learning in astrophysical research, particularly in predicting and understanding the intricate processes that govern star formation in various galactic environments. This approach can significantly enhance our comprehension of the universe's evolution.