Airline Passenger Satisfaction Prediction Using Supervised Machine Learning Algorithms: A Comparative Analysis of Operational and Service-Quality Factors

Abstract

Background: The airline industry produces large amounts of operational and customer-service data connected with passenger travel experience, operational efficiency, and service quality. Passenger satisfaction has become an important indicator of airline competitiveness because customer experience directly affects customer retention, airline reputation, and long-term business sustainability. Aims: The main aim of this study was to investigate whether supervised machine learning algorithms can effectively classify airline passenger satisfaction categories using operational and service-quality variables collected from airline-service environments. Study Design:  The study was designed as a quantitative supervised machine learning classification analysis involving comparative evaluation of multiple classification algorithms. Place and Duration of Study:  The experimental analysis was conducted using the Airline Passenger Satisfaction Dataset obtained from the Kaggle platform. The research and comparative analysis were completed between December 2025 to March 2026. Methodology: The dataset contained more than 120,000 passenger observations associated with demographic information, travel conditions, operational-flight variables, and customer-service evaluations. Before model implementation, preprocessing procedures including missing-value handling, categorical encoding, feature scaling, duplicate checking, and train-test splitting were applied to improve dataset quality and predictive stability. Five supervised classification algorithms were comparatively evaluated: XGBoost, CatBoost, LightGBM, Support Vector Machine, and AdaBoost. Model performance was assessed using Accuracy, Precision, Recall, F1-score, ROC-AUC analysis, and confusion matrix evaluation. Results: The experimental findings demonstrated that boosting-based ensemble models achieved stronger predictive performance compared with traditional classification approaches. XGBoost produced the highest overall classification consistency, while CatBoost and LightGBM also demonstrated reliable predictive capability within airline customer-service environments. XGBoost achieved the highest predictive performance with Accuracy above 96% and ROC-AUC value close to 0.98. Feature-importance analysis revealed that online boarding quality, seat comfort, inflight entertainment, operational delays, and digital-service accessibility represented the strongest variables influencing passenger satisfaction behavior. Demographic variables demonstrated comparatively lower predictive importance across classification models. Conclusion: The findings indicate that supervised machine learning approaches may provide useful support for airline customer analytics and passenger-satisfaction prediction systems. The study additionally demonstrates that airline passenger behavior is influenced by multidimensional interactions between operational reliability, digital-service accessibility, onboard comfort, and customer-service quality. Comparative evaluation of multiple machine learning algorithms may therefore support operational decision-making and customer-experience optimization within modern aviation-service environments.