Integrated Use of Data Mining and Statistical Analysis Methods to Analyze Air Traffic Delays
Linear regression is the primary data analysis method used in the development of air traffic delay models. When the data being studied does indeed have an underlying linear model, this approach would produce the best-fitting model as expected. However, it has been argued by ATM researchers [Wieland2005, Evans2004] that the underlying delay models are primarily non-linear. Furthermore, the delays being modeled often depend not only on the observable independent variables being studied but also on other variables not being considered. The traditional regression approach alone may not be best suited to study these type of problems. In this paper, we propose an alternate methodology based on partitioning the data using statistical and decision tree learning methods. We then show the utility of this model in a variety of different ATM modeling problems.