Data Analysis of Delays in Airline Networks

Cost-optimized airline resource schedules often imply a lack of delay tolerance in case of unforeseen disruptions, e.g. late check-ins, technical defects or airport and airspace congestion. Therefore, the consideration of timeliness and robustness has become an important topic in robust resource scheduling and a wide range of sophisticated scheduling approaches has been developed in recent years. However, these approaches depend on assumptions made concerning delay occurrences. A better understanding of delay mechanisms may lead to a better trade-off between cost-efficiency and robustness and is therefore the purpose of this paper. We provide a data-driven detection of decision rules for daytime delay trends, depending on spatio-temporal attributes. The focus is on interpretable rules whose prediction accuracy is compared to random forests as a non-parametric, automated modeling approach. The obtained results give an insight into both the nature of primary delay occurrence and the methodical potential of delay prediction in the context of robust resource scheduling.