The main goal of the study centers on developing a model for the purpose of air traffic forecasting by using off-the-shelf data mining and machine learning techniques. Although data driven modeling has been extensively applied in the aviation sector, little research has been done in the area of air traffic forecasting. This study is inspired by previous research focused on improving the Federal Aviation Administration (FAA) Terminal Area Forecasting (TAF) methodology, which historically assumed that the US air transportation system (ATS) network structure was static. Recent developments use data mining algorithms to predict the likelihood of previously un-connected airport-pairs being connected in the future, and the likelihood of connected airport-pairs becoming un-connected Despite the innovation of this research, it does not focus on improving the FAA's existing methodology for forecasting future air traffic levels on existing routes, which is based on relatively simple regression and growth models. We investigate different approaches for improving and developing new features within the existing data mining applications in air traffic forecasting. We focus particularly on predicting detailed traffic information for the US ATS. Initially, a 2-stage log-log model is applied to establish the significance of different inputs and to identify issues of endogeneity and multi-colinearity, while maintaining the simplicity of current models. Although the model shows high goodness of fit, it tested positive for both mentioned issues as well as presenting problems with causality. With the objective of solving these issues, a 3-stage model that is under development is introduced. This model employs logistic regression and discrete choice modelling. As part of future work, machine learning techniques such as clustering and neural networks will be applied to improve this model's performance.
展开▼
