Time-varying fields drive the motion of magnetic particles adsorbed on liquid drops due to interfacial constraints that couple magnetic torques to capillary forces. Such magneto-capillary particle dynamics and the associated fluid flows are potentially useful for propelling drop motion, mixing drop contents, and enhancing mass transfer between phases. The design of such functions benefits from the development and validation of predictive models. Here, we apply methods of Bayesian data analysis to identify and validate a dynamical model that accurately predicts the field-driven motion of a magnetic particle adsorbed at the interface of a spherical droplet. Building on previous work, we consider candidate models that describe particle tilting at the interface, field-dependent contributions to the magnetic moment, gravitational forces, and their combinations. The analysis of each candidate is informed by particle tracking data for a magnetic Janus sphere moving in a precessing field at different frequencies and angles. We infer the uncertain parameters of each model, criticize their ability to describe and predict experimental data, and select the most probable candidate, which accounts for gravitational forces and the tilting of the Janus sphere at the interface. We show how this favored model can predict complex particle trajectories with micron-level accuracy across the range of driving fields considered. We discuss how knowledge of this "best" model can be used to design experiments that inform accurate parameter estimates or achieve desired particle trajectories.