The adverse impact of flight delays on passenger satisfaction: An innovative prediction model utilizing wide & deep learning

• Published in: Journal of air transport management 2024-01, Vol.114, p.102511, Article 102511
• Main Authors: Song, Cen, Ma, Xiaoqian, Ardizzone, Catherine, Zhuang, Jun
• Format: Article
• Language: English
• Subjects: Feature interaction; Flight delay; Interpretability analysis; Passenger satisfaction prediction; Wide & deep model
• ISSN: 0969-6997; 1873-2089
• DOI: 10.1016/j.jairtraman.2023.102511

This article addresses the substantial negative influence of flight delays on passenger satisfaction and aims to bridge the research gap in understanding passenger satisfaction during delayed flights. We present a passenger satisfaction prediction model leveraging a real dataset from Kaggle. Through an examination of the interplay between individual in-flight services and passenger characteristics using the Pearson correlation coefficient and PCA-K-means clustering methods, we introduce a novel satisfaction prediction model built upon the deep learning Wide & Deep algorithm. Additionally, we employ the DeepLIFT algorithm to interpret the deep learning model and elucidate the salient features impacting passenger satisfaction, as revealed through feature importance analysis. Our findings demonstrate that the prediction model outperforms benchmark models such as MLP, SVM, and Random Forest, achieving higher accuracy. This study contributes to an enhanced comprehension of the multifaceted factors influencing passenger satisfaction following flight delays, and it offers valuable insights and recommendations for the enhancement of service quality among airline companies. •Constructing a Feature Interaction Based Model for Predicting Passenger Satisfaction on Board by Machine Learning Methods.•Improving the interpretability of models to better understand the needs and behaviors of passengers.•Using DeepLIFT algorithm for interpretability analysis, revealing the influencing factors of satisfaction prediction.•The prediction model shows higher accuracy compared to benchmark models such as MLP, SVM, and random forest.