Application of Data Mining to forecast Air Traffic: A 3-Stage Model using Discrete Choice Modeling

摘要

The main goal of this study centers on developing an aggregate air itinerary share model estimated at the city-pair level within the US air transportation system. This route demand assignment model is part of a new modeling approach that has as its ultimate output the prediction of detailed traffic information for the US air transportation system. In this approach, city-pair demand generation, route demand assignment and air traffic levels estimations are completed in 3 different stages within a single framework. Aiming to fully develop the overall model, in this paper we focus on estimating the 2nd stage, the air itinerary choice model. In order to achieve this, the first approach taken applies a multinomial logit model and uses a combination of stated preferences (SP) and revealed preferences (RP) data to estimate the model. By using a mixed dataset, we attempt to improve the RP model results, which often perform poorly due to high demand inelasticity. Preliminary results show the potential of this approach, although further analysis is required to understand the results obtained. For the final paper, different approaches and further interactions among the model attributes will be applied to improve the model’s performance.