Flight Time Prediction of Arrival Air Traffic Flows Using Time-Based Airspace Model Applying Machine-Learning Methods

Efficient airport operations are required as the demand for airports is expected to increase. In this study, we focused on arrival management in an air traffic system and attempted to predict flight times using machine learning. Results from our previous study indicate that it is essential to appropriately determine the airspace where the flight time is predicted in order to improve the prediction accuracy. Therefore, we propose the “time-based airspace” concept and apply it to a case study for the Tokyo International Airport. Time-based airspace is defined as airspace established based on an “equal time curve” calculated by clustering the positions of each aircraft at a certain time prior to arrival at the airport such that the average time required to arrive at the airport is equal. First, using the flight track data of arriving aircrafts, air traffic networks were constructed by applying kernel density estimation, and time-based airspaces were then designed based on the curves connecting each network node. Our proposed model results in a 20% reduction in variance of flight time compared to the previous “distance-based” prediction model and improves flight time prediction accuracy by 10%. Future prospects of our study include predictions using other methods, introducing other evaluation index, and further improving the airspace model.