Delay predictive analytics for airport capacity management

Abstract

Local delay predictions are crucial for optimizing airport capacity management, enhancing overall resilience, efficiency, and effectiveness of airport operations. This paper delves into the development and comparison of state-of-the-art predictive analytics techniques—spanning rule-based simulations, queuing models, and data-driven approaches—and demonstrates how they can empower informed decision-making toward mitigating the impact of potential delays across the whole spectrum of capacity management initiatives—from long-term strategic capacity planning to near real-time air traffic flow management. Using real-world data for four major airports in Southeast Asia, we comprehensively assess the performance of different methods and highlight the improved predictive capabilities achievable through data-driven methods and the incorporation of sophisticated features. Results show that (i) embedding queuing model features into machine learning models effectively captures congestion dynamics and nonlinear patterns, resulting in an improvement in predictive accuracy; (ii) incorporating advanced day-of features – lightning strikes, wind conditions, and propagated delays from prior hours – further enhances prediction accuracy, yielding $MAE$ gains ranging from 15% to 30%, contingent on the specific airport; (iii) in cases where limited information is available (years to months in advance of operations), conventional simulation and queuing models emerge as robust alternatives. Ultimately, we conceptualize and validate a delay prediction framework for airport capacity management, characterizing the different planning phases based on their specific delay prediction requirements and identifying appropriate methods accordingly. This framework offers practical guidance to airport authorities, enabling them to effectively leverage delay predictions into their airport capacity management practices.