Uncertainty Aware Column Generation for Crew Pairing Optimization Using Survival Analysis
Keywords: Reliable Scheduling and Planning, Column Generation, Machine learning for optimization, Survival Analysis, Crew pairing, Airline Operations Planning
Abstract: The crew pairing problem (CPP) is central to optimal planning and scheduling of operations in the airline industry, where the objective is to assign crews to cover a flight schedule at minimal cost while adhering to various logistical, personnel, and policy constraints. Despite the implementation of optimized schedules, operations are frequently disrupted by unforeseen events. This vulnerability stems from the deterministic nature of the CPP's base formulation, which fails to account for the uncertainties inherent in real-world operations. Existing solutions either aim to safeguard against a specified level of uncertainty or focus on worst-case scenarios. To this end, we propose a reliability-centric CPP formulation amenable to solution by column-generation (CG) SurvCG, that leverages survival analysis for dynamic quantification of uncertainty using the operation patterns in historical data. Applied to CPP, SurvCG forecasts and incorporates flight connection reliability into the optimization process. Through rigorous experiments on a large-scale first-of-its-kind real-world instance under regular and irregular operating conditions, we demonstrate that SurvCG achieves unprecedented improvements (up to 61%) over baseline in terms of total propagated delays, establishing SurvCG as the first data-driven solution for uncertainty-aware reliable scheduling.
Supplementary Material: zip
Primary Area: other topics in machine learning (i.e., none of the above)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8032
Loading