A Refined Model for Check-In and Security Staff Allocation at Airports Based on an Improved Bayesian Neural Network

Abstract

This study addresses the challenge of optimizing staff allocation in airport check-in and security areas by considering the inherent randomness and complexity of individual passenger characteristics, a factor often overlooked in traditional research. Unlike existing studies that treat passengers as a homogeneous group and rely on static models or monitoring data, this paper introduces a novel approach by leveraging passenger-specific characteristics to predict behavior and optimize staff allocation accordingly. The proposed model integrates Graph Neural Networks (GNN) within the genetic algorithm framework to enhance chromosome encoding, enabling more effective feature extraction and structure learning in Bayesian Neural Networks (BNN). The results show that the GNN-GA BNN model outperforms advanced models like DNN and XGBoost, improving accuracy by 7% and 2%, respectively, and enhancing AUROC by 3% and 4%. Additionally, the study introduces Conditional Rényi Entropy as a new metric for intervention analysis within BNNs, offering a more refined understanding of passenger behavior. A comprehensive passenger handling behavior prediction model and staff allocation strategy are developed, aiming to optimize staff requirements at airport terminals. A detailed case study further validates the model, demonstrating its ability to accurately predict passenger handling behavior and optimize staff allocation under varying conditions of gender, age, and luggage characteristics, which significantly impact service times.