Abstract—The basic objective of this study is to develop a model that analyzes and predicts the occurrence of flight arrival delays in the United States. Macroscopic and microscopic delay factors are discussed. In this research, we proposed new features that, to the best of our knowledge, were never used in previous studies, namely departure Part and Arrival Part of the day (Mornings, Afternoons, Evenings, Nights) and type of aircraft. U.S. domestic flight data for the year 2018, extracted from the Bureau of Transportation Statistics (BTS), were adopted in order to train the predictive model. We used efficient Machine Learning classifiers such as Decision Trees, K-Nearest Neighbors, Random Forest and Multilayer Perceptron. To overcome the issue of imbalanced data, sampling techniques were performed. We chose Grid Search technique for best parameters selection. The performance of each classifier was compared in terms of evaluation metrics, parameters tuning, data sampling and features selection. The experimental results showed that tuning and sampling techniques have successfully generated the best classifier which is Multilayer Perceptron (MLP) with an accuracy of 98.72% and a higher number of correctly classified flights.
 
Keywords—machine learning classification, flight delay prediction, multilayer perceptron, random forest, decision trees, k-nearest neighbors


Copyright © 2023 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.