Comparing Ensemble Learning Algorithms to Improve Flight Prediction Accuracy and Reliability

Flight delays are unavoidable, and they substantially affect the carriers’ profits and losses. Flight delays have a big impact on airline operations and airport on-time performance, both of which have a significant association with customer happiness. For airlines, estimating flight delays correctly is essential since the data may be used to boost client happiness and revenue for airline agencies. Many of the current methods for predicting flight delays are difficult to comprehend, rely on small samples of outdated data, and/or have limited or no scope for machine learning deployment. By analyzing the existing statistics for domestic flights within the United States of America (USA) for the years 2022–2023, this study creates a prediction model. Balancing imbalanced data using SMOTE is utilized to increase predictive performance while maintaining accuracy in class prediction for the minority class. To predict the flight delay, five ensemble classification models, namely AdaBoost classifier, bagging classifier, random forest classifier, extra trees classifier, and categorical boosting (CatBoost) classifier were employed. Exploratory data analysis is incorporated in the proposed model to shed light on the variables responsible for aircraft arrival delays and its causes. The CatBoost classifier outperformed other ensemble algorithms with an F1-score and AUC of 0.99 and 0.99, respectively.