AI Model Improves Airport Security Checkpoint Throughput Forecasting

This study introduces an innovative framework for accurately forecasting hourly passenger throughput at airport security checkpoints. The core of the approach involves transforming standard flight departure schedules into temporally aligned signals that represent passenger arrival intensity at security. This is achieved by distributing seat capacity across pre-departure hours using truncated Poisson kernels. These schedule-derived signals are then fed into a Temporal Fusion Transformer, alongside historical throughput data, scheduled activity, and temporal variables. The model demonstrated superior performance at Hartsfield-Jackson Atlanta International Airport, achieving a weighted mean absolute percentage error of 9.33% for direct six-hour forecasts, significantly better than recurrent neural network and long short-term memory models. The framework supports advanced staffing and operational planning without requiring individual passenger-flight matching.