There are many poorly-understood aspects of EP device physics. The impact of facility backpressure on Hall thruster operation, the anomalous collisionality seen within Hall thrusters, hollow cathodes, and magnetic nozzles, and the unsolved issue of plasma detachment from magnetic nozzles are all open questions which are unlikely to have closed-form, easily tractable expressions. The existence of these unsolved problems has prevented the development of fully-predictive thruster models. Data-driven methods allow us to make use of the large amount of available thruster data to approximate these unknown, non-classical effects and improve the predictive accuracy of existing models.