Flight Delay Prediction for Mitigation of Airport Commercial Revenue Losses Using Machine Learning on Imbalanced Dataset

Abstract

Flight delay is one of the factors that affect the decline in customer satisfaction and airport revenue. In addition to influencing customer perceptions of airport services, flight delay also has an impact on decreasing airport revenue and operation. This study models a flight delay prediction, and the process is carried out using Decision Tree, Random Forest, Gradient Boosted Tree, and XGBoost Tree algorithms. This study has also used and merged the weather characteristic data as secondary data to the airport operational flight data. To anticipate the imbalanced class, several sampling techniques were applied. Synthetic Minority Over Sampling Technique (SMOTE), Random Over-Sampling (ROS), Random Under-Sampling (RUS), and combining ROS with RUS are being used. The result of processing the analysis is in the form of a model to predict the category of flight delay. The model has been evaluated by using the Confusion Matrix and Area Under ROC Curve (AUC) value. The result of this study shows the Random Forest classifier with the combination of ROS + RUS technique and data split ratio of 90:10 gave the highest accuracy, error rate, and AUC value as shown as 82.58%, 17.42%, and 81.1% respectively on data testing. The result of the flight delay prediction model is expected to be a strategic recommendation for determining airport policies in the future. By implementing the best strategy related to the airport operation, it could carry out commercial planning in order to optimize airport commercial revenue.