http://arxiv.org/abs/2301.07143
The growing number of exoplanet discoveries and advances in machine learning techniques allow us to find, explore, and understand characteristics of these new worlds beyond our Solar System. We analyze the dataset of 762 confirmed exoplanets and eight Solar System planets using efficient machine-learning approaches to characterize their fundamental quantities. By adopting different unsupervised clustering algorithms, the data are divided into two main classes: planets with $\log R_{p}\leq0.91R_{\oplus}$ and $\log M_{p}\leq1.72M_{\oplus}$ as class 1 and those with $\log R_{p}>0.91R_{\oplus}$ and $\log M_{p}>1.72M_{\oplus}$ as class 2. Various regression models are used to reveal correlations between physical parameters and evaluate their performance. We find that planetary mass, orbital period, and stellar mass play preponderant roles in predicting exoplanet radius. The validation metrics (RMSE, MAE, and $R^{2}$) suggest that the Support Vector Regression has, by and large, better performance than other models and is a promising model for obtaining planetary radius. Not only do we improve the prediction accuracy in logarithmic space, but also we derive parametric equations using the M5P and Markov Chain Monte Carlo methods. Planets of class 1 are shown to be consistent with a positive linear mass-radius relation, while for planets of class 2, the planetary radius represents a strong correlation with their host stars’ masses.
M. Mousavi-Sadr, D. Jassur and G. Gozaliasl
Thu, 19 Jan 23
1/100
Comments: Submitted to MNRAS. 15 pages, 17 figures