Estimation of Delay Propagation in the National Aviation System Using Bayesian Networks

Flight delay creates major problems in the current aviation system. Methods are needed to analyze the manner in which micro-level causes propagate to create system-level patterns of delay. Traditional statistical methods are inadequate to the task. This paper proposes the use of Bayesian networks (BNs) to investigate and visualize propagation of delays among airports. The BN structure was developed from expert judgment and validated against empirical data. Parameters were estimated using a novel empirical Bayes approach in which regression estimates were used to construct a Dirichlet prior distribution, which was then updated from multinomial samples. Empirical results demonstrate greater predictive accuracy using our empirical Bayes approach than linear regression or Bayesian analysis with non-informative prior distributions. Our results clearly demonstrate the value of Bayesian networks for analyzing and visualizing how system-level effects arise from subsystem-level causes.