Machine learning algorithms and models for airport gate assignment problem: A systematic literature review

Abstract

As global air traffic continues to grow, the efficient utilization of airport terminal gates has become critical for adhering to turnaround schedules, minimizing arrival delay propagation, and reducing missed passenger connections. The Gate Assignment Problem (GAP)—which involves allocating arriving (and departing) aircraft to gates under operational constraints—has traditionally been addressed using exact optimization methods, heuristics, and metaheuristics. However, these methods struggle to either scale or adapt to the uncertainty and complexity of real-world airport operations. In recent years, Machine Learning (ML) has emerged as a promising alternative or complement to classical methods, offering a fundamentally data-driven approach to prediction and adaptive decision-making. ML techniques have shown potential to anticipate disruptions before they occur, rapidly approximate optimal solutions, and learn complex, nonlinear patterns in historical gate assignments that are difficult to codify using handcrafted heuristics. Yet, despite increasing academic interest, the application of ML to GAP remains fragmented and poorly synthesized. Existing studies apply diverse ML techniques and hybrid models but rarely benchmark them against traditional or standalone counterparts, and rely on inconsistent evaluation practices—using non-standardized, often proprietary datasets with limited reproducibility—hindering comparative analysis and generalizability.

This paper presents a systematic literature review (SLR) of ML-based approaches for solving the GAP, covering 21 peer-reviewed studies published between 2016 and 2025. We organize our review around three guiding research questions: (i) the comparative strengths and limitations of ML methods versus traditional optimization techniques; (ii) the design and performance of hybrid ML–optimization frameworks; and (iii) the types of datasets and feature sets used in ML-based GAP studies, and the extent to which they reflect the complexity and variability of real-world airport operations. Following the Kitchenham approach, we synthesize findings from peer-reviewed studies, highlighting trends and gaps to guide future gate assignment research and system development. Our review reveals that ML-based techniques—particularly reinforcement learning and supervised delay predictors—offer strong potential for handling uncertainty and improving decision quality compared to traditional optimization methods. However, their effectiveness is often limited by data availability and lack of interpretability. Hybrid ML–optimization frameworks show promise in combining predictive and search capabilities, but current designs are ad hoc and rarely benchmarked against their standalone components. Most ML-based GAP studies rely on narrow, single-airport datasets that omit key operational dynamics, limiting generalizability and real-world relevance. To address these gaps, we propose future directions: (1) developing robust and interpretable ML models that can adapt to changing operational contexts; (2) designing modular hybrid architectures that integrate feedback and support real-time updates; and (3) curating standardized multi-airport datasets—including gate occupancy records, passenger flows, ground operations, delay histories, and disruption events—for benchmarking and evaluation. Together, these steps can help transition ML-based GAP methods from academic prototypes to scalable, deployable tools for next-generation airport operations.