Enhancing Flight Delay Prediction Using Residual Neural Networks (ResNets)

Abstract

The most concern in airline sectors is flight delays because they have a big impact on airlines, passengers, and airports. This study used Residual Neural Network (ResNets), XGBoost, and LightGBM. The aim is to enhance flight delay prediction using ResNets. The performance of ResNets, which is a deep learning model, has been compared to the performances of XGBoost and LightGBM models, which are machine learning models. The dataset used is the domestic flights of United States from January 2019 to August 2023. The confusion matrix is used to make the comparisons between the selected models by summarizing prediction results, which are F1-score, accuracy, sensitivity, and precision. In addition, the validation and models' information, such as file size and prediction time, are used to assess the models' performance. The ResNets models have the best results, followed by LightGBM. The XGBoost has the worst results compared to other models.