This paper investigates the most important independent variables, including petrophysics and completion parameters, to estimate ultimate recovery with a machine-learning algorithm. A novel machine-learning model based on random forest regression is introduced to predict estimated ultimate recovery.