Date of Award
Dissertation - Open Access
Doctor of Philosophy in Aviation
College of Aviation
Dothang Truong, Ph.D.
First Committee Member
David Esser, Ph. D.
Second Committee Member
Bruce A. Conway, Ph.D.
Third Committee Member
David R. Bright, Ph.D.
This research sought to establish and utilize relationships between environmental variable inputs and airport efficiency estimates by data mining archived weather and airport performance data at ten geographically and climatologically different airports. Several meaningful relationships were discovered using various statistical modeling methods within an overarching data mining protocol and the developed models were tested using historical data. Additionally, a selected model was deployed using real-time predictive weather information to estimate airport efficiency as a demonstration of potential operational usefulness.
This work employed SAS® Enterprise Miner TM data mining and modeling software to train and validate decision tree, neural network, and linear regression models to estimate the importance of weather input variables in predicting Airport Arrival Rates (AAR) using the FAA’s Aviation System Performance Metric (ASPM) database. The ASPM database contains airport performance statistics and limited weather variables archived at 15-minute and hourly intervals, and these data formed the foundation of this study. In order to add more weather parameters into the data mining environment, National Oceanic and Atmospheric Administration (NOAA) National Centers for Environmental Information (NCEI) meteorological hourly station data were merged with the ASPM data to increase the number of environmental variables (e.g., precipitation type and amount) into the analyses.
Using the SAS® Enterprise Miner TM, three different types of models were created, compared, and scored at the following ten airports: a) Hartsfield-Jackson Atlanta International Airport (ATL), b) Los Angeles International Airport (LAX), c) O’Hare International Airport (ORD), d) Dallas/Fort Worth International Airport (DFW), e) John F. Kennedy International Airport (JFK), f) Denver International Airport (DEN), g) San Francisco International Airport (SFO), h) Charlotte-Douglas International Airport (CLT), i) LaGuardia Airport (LGA), and j) Newark Liberty International Airport (EWR). At each location, weather inputs were used to estimate AARs as a metric of efficiency easily interpreted by FAA airspace managers.
To estimate Airport Arrival Rates, three data sets were used: a) 15-minute and b) hourly ASPM data, along with c) a merged ASPM and meteorological hourly station data set. For all three data sets, the models were trained and validated using data from 2014 and 2015, and then tested using 2016 data. Additionally, a selected airport model was deployed using National Weather Service (NWS) Localized Aviation MOS (Model Output Statistics) Program (LAMP) weather guidance as the input variables over a 24-hour period as a test. The resulting AAR output predictions were then compared with the real-world AARs observed.
Based on model scoring using 2016 data, LAX, ATL, and EWR demonstrated useful predictive performance that potentially could be applied to estimate real-world AARs. Marginal, but perhaps useful AAR prediction might be gleaned operationally at LGA, SFO, and DFW, as the number of successfully scored cases fall loosely within one standard deviation of acceptable model performance arbitrarily set at ten percent of the airport’s maximum AAR. The remaining models studied, DEN, CLT, ORD, and JFK appeared to have little useful operational application based on the 2016 model scoring results.
Scholarly Commons Citation
Maxson, Robert William, "Prediction of Airport Arrival Rates Using Data Mining Methods" (2018). PhD Dissertations and Master's Theses. 419.