Predicting Monthly Flight Cancellations in The Post-Pandemic Times Using Machine Learning Methods

Abstract

This study explores the trends of flight cancellations using machine learning techniques in the post-pandemic times of COVID-19. The study identifies important factors influencing flight cancellations. The techniques used are linear regression, ridge regression, gradient-boosted regression, decision forest, and decision tree regression. Monthly data of ten leading US airlines with 33536 patterns of flight cancellations has been considered. The results of these sophisticated methods can help to make informed decisions in advance. The best training score of 0.99 is corresponding to the decision forest, followed by the random forest with a score of 0.93. Thus, Decision Forest has learned the data very well. Additionally, the best testing score of 0.65 is corresponding to a random forest, followed by a decision tree with a score of 0.37. Thus, on the basis of training decision tree is the best, whereas on the basis of testing, random forest is the best model. From the results, it is evident that monthly arrivals canceled have a relationship of 0.21 with both arrival delay15 and number of flights delayed attributed to the national aviation system. Moreover, its relationships with delayed flights attributable to late aircraft, weather and security reasons are 0.18, 0.12 and 0.07, respectively. The major contributions of the research are twofold. In the first place, predicting the number of cancellations in the post-pandemic circumstances. Next, machine learning techniques are implemented to draw meaningful conclusions for managing future airline schedules. Thus, the issue of flight cancellation can be properly managed.