Ground Delay Program Analytics with Behavioral Cloning and Inverse Reinforcement Learning

Ground Delay Program Analytics with Behavioral Cloning and Inverse Reinforcement Learning. Bloem, M. & Bambos, N. Journal of Aerospace Information Systems, 12(3):299–313, 2015.

Paper doi abstract bibtex

Historical data are used to build two types of models that predict Ground Delay Program implementation decisions and produce insights into how and why those decisions are made. More specifically, behavioral cloning and inverse reinforcement learning models are built that predict hourly Ground Delay Program implementation at Newark Liberty International and San Francisco International airports. Data available to the models include actual and scheduled air traffic metrics and observed and forecasted weather conditions. The developed random forest models are substantially better at predicting hourly Ground Delay Program implementation for these airports than the developed inverse reinforcement learning models. However, all of the models struggle to predict the initialization and cancellation of Ground Delay Programs. The structure of the models are also investigated in order to gain insights into Ground Delay Program implementation decision making. Notably, characteristics of both types of model suggest that Ground Delay Program implementation decisions are more tactical than strategic: they are made primarily based on conditions now or conditions anticipated in only the next couple of hours.

@article{bloem_ground_2015,
	title = {Ground {Delay} {Program} {Analytics} with {Behavioral} {Cloning} and {Inverse} {Reinforcement} {Learning}},
	volume = {12},
	url = {https://doi.org/10.2514/1.I010304},
	doi = {10.2514/1.I010304},
	abstract = {Historical data are used to build two types of models that predict Ground Delay Program implementation decisions and produce insights into how and why those decisions are made. More specifically, behavioral cloning and inverse reinforcement learning models are built that predict hourly Ground Delay Program implementation at Newark Liberty International and San Francisco International airports. Data available to the models include actual and scheduled air traffic metrics and observed and forecasted weather conditions. The developed random forest models are substantially better at predicting hourly Ground Delay Program implementation for these airports than the developed inverse reinforcement learning models. However, all of the models struggle to predict the initialization and cancellation of Ground Delay Programs. The structure of the models are also investigated in order to gain insights into Ground Delay Program implementation decision making. Notably, characteristics of both types of model suggest that Ground Delay Program implementation decisions are more tactical than strategic: they are made primarily based on conditions now or conditions anticipated in only the next couple of hours.},
	number = {3},
	urldate = {2017-11-02},
	journal = {Journal of Aerospace Information Systems},
	author = {Bloem, Michael and Bambos, Nicholas},
	year = {2015},
	keywords = {MachineLearning},
	pages = {299--313},
}

Downloads: 0

{"_id":"3vLfaYKTRxPseA3LM","bibbaseid":"bloem-bambos-grounddelayprogramanalyticswithbehavioralcloningandinversereinforcementlearning-2015","author_short":["Bloem, M.","Bambos, N."],"bibdata":{"bibtype":"article","type":"article","title":"Ground Delay Program Analytics with Behavioral Cloning and Inverse Reinforcement Learning","volume":"12","url":"https://doi.org/10.2514/1.I010304","doi":"10.2514/1.I010304","abstract":"Historical data are used to build two types of models that predict Ground Delay Program implementation decisions and produce insights into how and why those decisions are made. More specifically, behavioral cloning and inverse reinforcement learning models are built that predict hourly Ground Delay Program implementation at Newark Liberty International and San Francisco International airports. Data available to the models include actual and scheduled air traffic metrics and observed and forecasted weather conditions. The developed random forest models are substantially better at predicting hourly Ground Delay Program implementation for these airports than the developed inverse reinforcement learning models. However, all of the models struggle to predict the initialization and cancellation of Ground Delay Programs. The structure of the models are also investigated in order to gain insights into Ground Delay Program implementation decision making. Notably, characteristics of both types of model suggest that Ground Delay Program implementation decisions are more tactical than strategic: they are made primarily based on conditions now or conditions anticipated in only the next couple of hours.","number":"3","urldate":"2017-11-02","journal":"Journal of Aerospace Information Systems","author":[{"propositions":[],"lastnames":["Bloem"],"firstnames":["Michael"],"suffixes":[]},{"propositions":[],"lastnames":["Bambos"],"firstnames":["Nicholas"],"suffixes":[]}],"year":"2015","keywords":"MachineLearning","pages":"299–313","bibtex":"@article{bloem_ground_2015,\n\ttitle = {Ground {Delay} {Program} {Analytics} with {Behavioral} {Cloning} and {Inverse} {Reinforcement} {Learning}},\n\tvolume = {12},\n\turl = {https://doi.org/10.2514/1.I010304},\n\tdoi = {10.2514/1.I010304},\n\tabstract = {Historical data are used to build two types of models that predict Ground Delay Program implementation decisions and produce insights into how and why those decisions are made. More specifically, behavioral cloning and inverse reinforcement learning models are built that predict hourly Ground Delay Program implementation at Newark Liberty International and San Francisco International airports. Data available to the models include actual and scheduled air traffic metrics and observed and forecasted weather conditions. The developed random forest models are substantially better at predicting hourly Ground Delay Program implementation for these airports than the developed inverse reinforcement learning models. However, all of the models struggle to predict the initialization and cancellation of Ground Delay Programs. The structure of the models are also investigated in order to gain insights into Ground Delay Program implementation decision making. Notably, characteristics of both types of model suggest that Ground Delay Program implementation decisions are more tactical than strategic: they are made primarily based on conditions now or conditions anticipated in only the next couple of hours.},\n\tnumber = {3},\n\turldate = {2017-11-02},\n\tjournal = {Journal of Aerospace Information Systems},\n\tauthor = {Bloem, Michael and Bambos, Nicholas},\n\tyear = {2015},\n\tkeywords = {MachineLearning},\n\tpages = {299--313},\n}\n\n\n\n\n\n\n\n","author_short":["Bloem, M.","Bambos, N."],"key":"bloem_ground_2015","id":"bloem_ground_2015","bibbaseid":"bloem-bambos-grounddelayprogramanalyticswithbehavioralcloningandinversereinforcementlearning-2015","role":"author","urls":{"Paper":"https://doi.org/10.2514/1.I010304"},"keyword":["MachineLearning"],"metadata":{"authorlinks":{}},"html":""},"bibtype":"article","biburl":"http://bibbase.org/zotero/LordMzn","dataSources":["g5jeJwLDDqzZ7QHm5"],"keywords":["machinelearning"],"search_terms":["ground","delay","program","analytics","behavioral","cloning","inverse","reinforcement","learning","bloem","bambos"],"title":"Ground Delay Program Analytics with Behavioral Cloning and Inverse Reinforcement Learning","year":2015}