Researchers have a better way to predict flight delays
Reserachers at State University of New York (SUNY) Binghamton have discovered a new way to predict flight delays using an Artifical Neural Network (ANN), an “interconnected group of computerized nodes that work together to analyze a variety of variables to estimate an outcome.” In this case, there are 14 variables, some of which include day of the week, origin airport, weather and security. Integrating these variables lead to the creation of a schedule of safe landing times for each airplane that has not landed at that moment. What’s novel about this approach is the number of variables necessary to create a schedule. Previous models took more time to complete due to the number of variables necessary to schedule safe landing times. Hence, many airplanes would land before an output of accurate schedules for safe landing times would be released. The use of an ANN has predicted the length of delays with about 20 percent more accuracy than traditional models and required about 40 percent less time to arrive at those conclusions.
The ANN is a complicated version of the networks we have discussed in class. Each node (in this case, a node is referred to as a neuron) is given a weight, 0 or 1. A 0 means that the neural network is off while a 1 means it is active. In this case the neurons or nodes are variables like the ones described in the paragraph above. A binary system of nodes (neurons) is employed to give equal weight to nominal variables that cannot be processed in a traditional ANN model (ex: day of the month, 25 must be given the same weight as 31). For other variables, like the origin of departure use a layered binary approach. Activation of different sublayers in the binary system will give different information about the origin of departure for each flight.
The research team at SUNY Binghamton plans to test this hypothesis at other airports, as this specific study took place at JFK Airport. To expand to real life applications besides predicting flight times, the team plans to include degrees of truth, rather than a layered binary (true/false) system to the ANN. This means less working memory would be used to calculate information, which would lead to quicker response times for the program.
http://www.sciencedirect.com/science/article/pii/S1877050916324942
https://www.sciencedaily.com/releases/2016/11/161114103905.htm