A flexible modeling approach for the accurate approximation of convection-dominated reactive-species boundary layers is introduced. A substitute problem is solved numerically and analyzed by employing statistical methods. The numerical data are then used to train a machine learning model that can be used to approximate the reactive mass transfer locally if a direct resolution of the concentration boundary layer is infeasible. Compared to previous modeling approaches, the machine learning model replaces the analytical solution of a simplified substitute problem, which makes it applicable to more complicated and general settings.