Analyzing the impact of COVID-19 on flight cancellation using machine learning and deep learning algorithms for a highly unbalanced dataset

Abstract

Flight cancellations can be caused by many factors, including adverse weather conditions, and can result in lost money and time, etc. The COVID-19 pandemic has significantly exacerbated this situation, leading to a significant decrease in air travel. In 2020, the number of cancelled flights increased by 200% over 2019 and the number of flights decreased by 38%. This research focused on analyzing the impact of COVID-19 on flight cancellation using publicly available datasets from different locations. We looked further into the impact of class imbalance and techniques to reduce its effects on classification errors. The research was performed using four data sets, six re-sampling techniques, and 12 modeling algorithms. Random oversampling combined with random subsampling outperformed all other resampling techniques and multi-layer perceptron (MLP) was the best among all other machine learning models. For validation, we used the same resampling technique to two additional datasets namely income and diabetes datasets. The results showed that combining random oversampling with subsampling improved the accuracy of machine learning models.