PREDICTIVE MODELS OF THE NATIONAL AIRSPACE SYSTEM USING BAYESIAN NETWORKS

This paper focuses on a technique to model the National Airspace System which does not require selection of reference data and provides a framework to discover the complicated effects of congestion in one region of airspace on the operation of another. Since the interactions at this level are complex, only historical data are used in the analysis. From these data, a Bayesian Network model of the system is learned that is capable of predicting the number of aircraft in certain regions of the airspace at a given time with greater accuracy than similar linear regression models. The model is also used to automatically determine which regions of the system are the most critical to overall performance. The model is also used to automatically rank the regions of the system based on how important congestion in each region is to the overall performance of the system.