Data-driven modeling and analysis of the U.S. air transportation network and its resilience to extreme events

This thesis presents a data-driven approach for analyzing and predicting delays of an air transportation network using publicly available data. The first part of this thesis details methods to quantify the resilience of the network. Traditionally, network metrics rely on removal of nodes heuristically to measure the resilience of the network. We propose two new approaches that rely on statistical measures to quantify the resilience of the network based on historical data. Data-driven analysis of the network's resilience based on these metrics enables comparison and implementation in the real-world. The second half of this thesis details development of a neural network model that can predict future delays in a network based on past and current conditions. Previous work using this approach has shown the ability to predict delays based on temporal, weather or network metrics. This work shows a method to build prediction models by combining temporal, network-level features, congestion, and weather related data. As part of this approach, we devised a new metric that reduces the dimensionality of network-level information into a single variable. Finally, we compare the performance of the neural network by changing the hyperparameters for optimal performance