Traffic flow on networks requires, in particular, knowledge on the behavior across traffic intersections. For macroscopic models based on hyperbolic conservation laws there exist nowadays many ad-hoc models describing this behavior. Based on car trajectory data we propose a novel framework combining data-fitted models with the requirements of consistent coupling conditions for the macroscopic models of traffic junctions. A method for deriving corresponding density and flux of the traffic close to the junction for data-driven models is presented. Within those models parameter fitting as well as machine-learning approaches enter to obtain suitable boundary conditions for macroscopic first and second-order traffic flow models. The prediction are compared with existing traffic flow models at the junction. Numerical results on the data are presented.