Predicting the orbifold origin of the MSSM

MSSM-like string models from the compactification of the heterotic string on toroidal orbifolds (of the kind $T^6/P$) have distinct phenomenological properties, like the spectrum of vector-like exotics, the scale of supersymmetry breaking, and the existence of non-Abelian flavor symmetries. We show that these characteristics depend crucially on the choice of the underlying orbifold point group $P$. In detail, we use boosted decision trees to predict $P$ from phenomenological properties of MSSM-like orbifold models. As this works astonishingly well, we can utilize machine learning to predict the orbifold origin of the MSSM.