A Two-Stage Machine Learning and Metaheuristic Framework for Airport Gate Conflict Detection and Reassignment

Gate assignment is considered one of the most important tasks in airport opera- tions. Poor gate allocation may result in delay propagation, increased operational cost, and degraded passenger experience. The early and accurate identification of the risk of a potential conflict between two flights for the same gate allows for timely reassignment and more reliable day-of-operations planning. Motivated by the need for objective and scalable assessment of gate conflicts, this work moves one step further towards a rapid, automated, and reliable computation frame- work for managing airport gates. This paper presents a comparison between two optimization strategies, Tabu Search versus an Adaptive Genetic Algorithm, on top of a learned conflict prediction model, to determine suitable approaches for real-time and batch gate optimization, respectively. In the proposed system, a two-stage pipeline has been utilized to accomplish comprehensive conflict management. In the first stage, an XGBoost classifier predicts the gate conflict probability for an incoming or scheduled flight based on historical patterns of traffic and contextual features. Stage 2 involves trans- mitting high-risk flight paths to an optimal layer, wherein Tabu Search offers fine-grained and flight-by-flight reassignment methods involving real-time appli- cations, and an Adaptive Genetic Algorithm respectively tackles larger batches of flight paths to optimize global scheduling. Real-life flight and gate data analysis has revealed that Stage 1 provides reliable conflict predication results adequate for real-time screening, and all methods applied in Stage 2 manage to lower predicted conflict probabilities by at least 70% when reassigning flight paths. Although Tabu Search generates responses in no time when dealing with indi- vidual flight paths, the Genetic Algorithm succeeded in providing considerable flight-level conflict prevention at relatively lower overall latency times, thereby confirming that the developed multi-stage model using machine learning pred- ication and meta-search algorithms offers great potential in being adopted in airport-related decision-support tools striving to provide optimal results in min- imizing potential gate-related flight conflicts and enhance overall operational elasticity