This paper presents an approach for predicting delay states of airports in the United States Air Transportation Network using publicly available data. We illustrate a procedure to predict future airport delays in the network based on temporal, network-level, congestion, and weather-related features from past and current data. As part of this approach, we devised a network delay metric that reduces the dimensionality of network-level delay information into a single variable, thus reducing the feature space and enabling use of classic statistical models. We consider two model types for this paper: a Neural Network model and a Logistic Regression model. We find that prediction performance is most significantly impacted by forecast interval and delay threshold for the presented cases. Similar test accuracies are seen among considered models, with accuracies ranging from 59.5% to 95.8% depending on problem settings. We also test performance of a Neural Network model for the difficult task of predicting airport delay states during extreme events, and find a test accuracy of 69.2% for data from Hurricane Harvey in 2017.