Towards Carrier-Level Airline Passenger Demand Forecasting: A Hierarchical Attention Framework for Strategic Planning

Abstract

Airlines face increasingly complex capacity planning decisions requiring passenger demand forecasts at unprecedented granularity. We introduce a hierarchical framework generating carrier-specific passenger forecasts across 10-year strategic horizons. The framework combines attention-LSTM route forecasting with persistence-based carrier decomposition, achieving 9.55% MAPE on route-level predictions while outperforming ARIMA by 19.6% (p < 0.001). Validation on 25 years of T-100 data (10.26M observations) across five major U.S. routes demonstrates the framework maintains zero- sum constraints (shares sum to 100% monthly) while preserving competitive structures: hub-dominated markets maintain 65-70% leaders, fragmented markets continue balanced competition. The framework generates 120-month forecasts (2025-2034) for 26 carrier-route combinations, providing air-lines with operational tools for fleet deployment and capital allocation decisions.