Air traffic generators are widely used in the air traffic control and aeronautics industry for training purposes and to validate new concepts or new systems. But it is difficult for a machine to generate realistic trajectories and behavior for an aircraft which lead to the participation of many humans in the simulation to compensate. Traditional generators require performance data for aircrafts and heavy preparation to generate realistic trajectories. Our approach takes advantage of widespread flight location data to learn from their behavior and improve traffic generation. Combining cooperative multi-agent learning methods [1] and big data we show that it is possible to design and generate a realistic and intelligent traffic more easily and without human intervention.