George C. Angelou, Saskia Hekker, Sarbani Basu, Warrick Ball, Elisabeth Guggenberger

We use machine learning to build a constrained multiple regression model for rapidly estimating the fundamental stellar parameters of main-sequence solar-like stars (Bellinger & Angelou 2016). We train a random forest of decision trees with scikit-learn on a matrix of stellar models generated with MESA that we varied quasi-randomly in mass, initial helium and metallicity abundances, mixing length, the strength of convective overshooting, and the efficiency of gravitational settling. We additionally compute frequencies of each stellar model using ADIPLS and summarize them to obtain averaged large and small frequency separations and frequency ratios. We supply the global asteroseismic properties and other observable quantities of our stellar models to the random forest algorithm and produce a statistical model relating observable quantities to fundamental stellar parameters. We validate this technique on a hare-and-hound exercise and the Sun, and then apply it to 16 Cyg A & B and finally 34 planet-hosting candidates.