Perpustakaan Digital ITB

Airlines increasingly rely on digital self-service technologies, including online booking, mobile check-in, and digital boarding, as part of the passenger journey. However, passenger dissatisfaction may arise from a combination of digital service, cabin experience, passenger profile, and operational factors. This undergraduate thesis examines airline passenger dissatisfaction using an explainable machine learning approach, with Online Boarding selected as the primary analytical case. In a preliminary Random Forest analysis, Online Boarding showed the highest feature importance in relation to Overall Satisfaction, motivating its selection for further analysis. Drawing on the Airline Passenger Satisfaction dataset comprising 129,880 passenger records, the study converts the Online Boarding score into a binary target: scores of 1–2 are coded as dissatisfaction, while scores of 3–5 are coded as non-dissatisfaction. Logistic Regression serves as a baseline, while XGBoost is used as the main predictive model. SHAP is then used to identify the features that contribute most strongly to model predictions and the direction of their contributions. The tuned XGBoost model achieves 0.9380 accuracy, 0.9849 AUC, an 0.8969 F1-score for the dissatisfied class, and 0.9317 balanced accuracy. SHAP analysis identifies Inflight WiFi Service, Ease of Online Booking, Seat Comfort, passenger class, and customer loyalty status among the strongest contributors to predicted dissatisfaction. Cross-tabulation further shows that unfavorable service ratings vary across passenger segments and contexts. Overall, the findings indicate that airline passenger dissatisfaction is associated with multiple aspects of the passenger experience, with Online Boarding serving as the primary analytical case. The study provides interpretable insights that can support targeted service improvement across different passenger segments.