Rolling-Calibrated Conformal Decision Support for Seasonal Airport Taxi-Time Planning under Distribution Shift

Abstract

Seasonal airport taxi-time planning is considered to be an asymmetric-risk decision problem, under-planning in operations can propagate delay and disrupt airline schedules, whereas excessive padding can reduce aircraft utilization and contribute to schedule creep which explains the importance of smart balance. This paper formulates seasonal taxi-time planning as a one-sided risk-control problem and studies a rolling-calibrated conformal decision-support framework for airline scheduling under non-stationary operating conditions. The proposed framework combines a pooled seasonal point forecaster with rolling one-sided conformal calibration to convert point predictions into planning recommendations with an explicit marginal under-planning target. This paper operationalizes existing one-sided conformal calibration for a real aviation planning problem, evaluates its empirical behavior over 17 IATA seasons and 722 airports in the EUROCONTROL CODA dataset, and provides deployment-oriented diagnostics for airport-level monitoring, cohort instability, sparse-history airports, and wake-turbulence-category heterogeneity.Results show that traditional persistence heuristics can fail during the disruption/recovery period, whereas rolling conformal calibration substantially reduces under-planning toward the nominal planning target, at the cost of additional padding. We explicitly distinguish formal conformal guarantees under exchangeability from empirical rolling adaptation under distribution shift. The framework is therefore best interpreted as an auditable, human-supervised decision-support layer for seasonal planning, not as an autonomous scheduling rule or a guarantee-restoration mechanism under arbitrary drift.