Queue behavioural patterns for passengers at airport terminals: A machine learning approach

Passengers go through different handling processes inside airport terminal buildings. The quality of these processes is usually measured by the time passengers require and by the level of comfort experienced by them. We present an analysis of behavioural patterns in queues at check-in desks and security controls, which are two of the most critical processes regarding passenger service. The passengers' flow is simulated to obtain queue lengths at one busy European airport between 2014 and 2016, supported by real flight data. Simulation is designed as a store-and forward cell-based system, whose parameters have been tuned and validated with real data from observations and empirical capacity and demand studies within the airport. Random Forest algorithms are then implemented to develop different models for each parameter prediction, after a data analysis stage based on statistical and visualization methods. Feature analysis techniques between dependent variables and the target outputs (queue lengths) determine which are the fundamental elements to explain queue behaviour and to predict target variables. We provide a method to forecast behavioural patterns at check-in desks and security controls, to help airport operators to implement adequate response policies. Queue behavioural patterns are captured by Machine Learning models, which can be used to offer improved passenger services (such as real-time predictions for expected waiting time at queues), or can be considered in a dynamic approach for terminal services design (as the entire progress of terminal handling depends on the stochastic behaviour of passengers). This could be a key tool for managing passengers demand and optimise the infrastructure's capacity through resource allocation.